Journal of Agriculture and Food Research 14 (2023) 100880

Contents lists available at ScienceDirect

Journal of Agriculture and Food Research

journal homepage: [Link]/journal/journal-of-agriculture-and-food-research

Machine learning enabled IoT system for soil nutrients monitoring and
crop recommendation
Md Reazul Islam a, Khondokar Oliullah a, Md Mohsin Kabir a, Munzirul Alom a, M.F. Mridha b, *
a
Department of Computer Science and Engineering, Bangladesh University of Business and Technolog, Rupnagar R/A, Mirpur-2, Dhaka, 1216, Bangladesh
b
Department of Computer Science, American International University-Bangladesh, Dhaka, 1229, Bangladesh

A R T I C L E I N F O A B S T R A C T

Keywords: Agriculture plays a vital role in feeding the growing global population. But optimizing crop production and
Machine learning resource management remains a significant challenge for farmers. This research paper proposes an innovative
Internet of things ML-enabled IoT device to monitor soil nutrients and provide accurate crop recommendations. The device utilizes
Agriculture
the FC-28 sensor, DHT11 sensor, and JXBS-3001 sensor to collect real-time data on soil composition, moisture,
Crop recommendation
Soil nutrients
humidity, temperature, and for nutrient levels. The collected data is transmitted to a server using the MQTT
Sensor protocol. Machine learning algorithms are employed to analyze the collected data and generate customized
Cloud computing recommendations, including a possible high-yielding crop list, fertilizer names, and its amount based on crop
Crop quality assessment requirements and soil nutrients. Furthermore, the applied fertilizers and treatments to the field during pro­
duction are stored in the database. As a result, it has become possible to determine the quality of the produce at
the consumer level through the mobile app. The system’s effectiveness is evaluated through field experiments,
comparing its performance with traditional methods. The results demonstrate the device’s ability to enhance
crop productivity and optimize resource utilization, promoting sustainable agricultural practices and food se­
curity. The research contributes to IoT-enabled agriculture, demonstrating the potential of ML techniques in
improving soil nutrient management, facilitating informed decision-making about crop fertilizers, and assessing
the quality of produced crops at the consumer level.

1. Introduction real-time, accurate data on soil nutrient levels and generate customized
recommendations based on the specific requirements of different crops.
The advent of the Internet of Things (IoT) and the rapid advance­ This research paper proposes an innovative ML-enabled IoT device
ments in machine learning (ML) techniques, ranging from healthcare to that integrates advanced sensor technologies, data analytics algorithms,
transportation [2,11,20,24,25,34,36,41,50] have opened up new ave­ and crop science knowledge to monitor soil nutrients and provide
nues for improving agricultural practices. In recent years, there has been intelligent crop recommendations. The proposed device harnesses the
a growing emphasis on leveraging these technologies to address the power of ML algorithms to analyze real-time data collected from sensors
challenges faced by farmers in optimizing crop production and reducing embedded in the soil, including moisture, ph, temperature, humidity,
resource wastage [33]. Among these challenges, the effective manage­ and nutrient sensors. These sensors continuously measure and transmit
ment of soil nutrients and accurate crop recommendations [47] play vital information about the soil composition, NPK, temperature, hu­
pivotal roles in ensuring sustainable and efficient agricultural practices. midity, and moisture content to a central cloud, where the ML algo­
Traditional methods of soil nutrient analysis and crop recommen­ rithms process the data to extract actionable insights.
dation have often been labor-intensive, time-consuming, and rely on The ML algorithms employed in this device utilize sophisticated data
subjective assessments. As a result, farmers are often left with subopti­ modeling techniques, such as random forest, logistic regression, LGBM,
mal decisions, leading to inefficient use of resources, reduced yields, and and neural networks, to identify and understand the intricate relation­
increased environmental impact [10]. To overcome these limitations, ships between soil nutrient levels, environmental factors, and crop re­
there is a pressing need for innovative solutions that can provide quirements. By leveraging historical data, the ML algorithms are trained

* Corresponding author.
E-mail addresses: [Link]@[Link] (M.R. Islam), [Link]@[Link] (K. Oliullah), mdmkabi@[Link] (M.M. Kabir), cse.alomb008@[Link]
(M. Alom), [Link]@[Link] (M.F. Mridha).

[Link]
Received 29 June 2023; Received in revised form 1 November 2023; Accepted 14 November 2023
Available online 23 November 2023
2666-1543/© 2023 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license ([Link]
nc-nd/4.0/).
M.R. Islam et al. Journal of Agriculture and Food Research 14 (2023) 100880

to generate accurate predictions and recommendations tailored to spe­ resulting in improved crop yield and water use efficiency.
cific crops, taking into account their nutrient demands, growth stages, [49] proposes a system that utilizes an Internet of Things (IoT)
and environmental conditions. Additionally, this system can show the platform for monitoring citrus moisture, temperature, and nutrient
quality of the produce at the consumer level when it will be in the levels to provide early warning systems and decision-making support for
market. The outcomes of this research endeavor have the potential to large-scale citrus cultivation. The system employs a single-point
revolutionize the way farmers manage soil nutrients and make crop multi-layer detection method to obtain temperature, humidity, and
decisions. By providing real-time, accurate data on soil conditions and nutrient data, which expands the detection range and improves accu­
customized crop recommendations, the proposed ML-enabled IoT device racy. The system uses ARM11-based S3C6410 as the master chip.
aims to empower farmers with the tools and knowledge necessary to Additionally, an expert knowledge base tailored to the Three Gorges
enhance productivity, optimize resource utilization, food security, and reservoir area is established, offering guidance for citrus fertilization
promote sustainable agricultural practices. and irrigation decision support based on local geographical conditions
Contributions to this work include. and citrus management experience. Field testing demonstrates that the
system enables scientific management decisions based on citrus growth
a) The design and development of an IoT device with integrated ma­ conditions. However, the system still has limitations, such as fragmented
chine learning capabilities for continuously monitoring soil mois­ solutions and high costs, which need to be addressed.
ture, humidity, temperature, and NPK levels using three different Machine learning techniques have shown great potential in
sensors. This device provides real-time data, enabling precise and analyzing soil data and predicting nutrient levels [15,16,32]. ML algo­
timely agricultural decision-making. rithms can identify patterns and correlations in large datasets, enabling
b) The application of machine learning algorithms to analyze the accurate predictions and proactive decision-making. In a study by
collected data and generate crop recommendations based on Ref. [8], an ML model was developed to predict soil nutrient levels based
observed patterns. These recommendations contribute to optimizing on historical data, environmental factors, and crop requirements. The
crop growth and maximizing yield potential, promoting efficient model achieved high accuracy in predicting nutrient deficiencies and
resource utilization in agriculture. excesses, aiding in the timely adjustment of fertilizer application rates.
c) Thorough evaluation of the developed device through field experi­ [1] proposes a system to predict three essential soil fertility elements
ments, including a comparative analysis with existing methods. This (OM, K2O and P2O5) using three machine learning approaches like
assessment highlights the capabilities and advantages of the device multiple linear regression (MLR), support vector machine (SVM) and
over traditional approaches, demonstrating its potential to revolu­ random forest (RF). They collets 400 soils from Doukala, central
tionize agricultural practices. Morocco. According to the findings, texture, carbonates, and cation
d) The development of a mobile application to engage end-users and exchange capacity were the key factors that significantly influenced the
facilitate crop quality assessment. By providing access to the prediction of OM, P2O5, and K2O. They also proposed fertilizer
collected data, consumers can monitor the quality of the crops pro­ recommendation model.
duced. This feature enhances transparency in the agricultural supply Crop recommendation systems play a critical role in guiding farmers
chain, empowering consumers to make informed decisions about the towards optimal crop selection based on soil conditions and market
agricultural produce they consume. demands. ML algorithms have been successfully employed to develop
crop recommendation models that consider various factors, such as soil
The paper is organized as follows: Section 2 comprehensively re­ nutrients, climate conditions, and market trends [4,6,9,27,37]. For
views related work. Section 3 describes the methodology and materials instance Ref. [46], proposed a crop recommendation system that uti­
used in the study, including IoT devices and sensors, data transmission lized ML algorithms to analyze historical crop yield data, weather pat­
and analysis, machine learning model design and implementation, and terns, and soil nutrient levels. The system provided personalized crop
the proposed framework. Section 4 presents the analysis of the results, recommendations to farmers, resulting in improved crop productivity
including data collection and preprocessing, evaluation metrics, training and profitability.
and testing of the proposed model, and performance analysis. Section 5 The integration of IoT and ML technologies offers a comprehensive
provides an in-depth discussion of the study’s findings and limitations. approach to soil nutrients monitoring and crop recommendation [26,28,
Finally, Section 6 presents the conclusion and future directions for 40,45]. By combining real-time soil data collected through IoT devices
research in this field. with ML algorithms, accurate predictions and customized recommen­
dations can be generated. For instance Ref. [18], developed an
2. Related works ML-enabled IoT device that employed various soil sensors and ML al­
gorithms to monitor soil conditions and provide crop-specific recom­
The management of soil nutrients and accurate crop recommenda­ mendations. The device successfully improved crop yields and
tions are crucial for sustainable and efficient agricultural practices. optimized resource utilization.
Traditional methods of soil nutrient analysis and crop decision-making A system that consists of sensors, an Arduino board, and an Amazon
often rely on subjective assessments and labor-intensive processes, Web Service has been presented. The used machine learning methods,
resulting in suboptimal resource utilization and reduced yields. How­ the cloud, and a mobile application to analyze and visualize data. This
ever, the integration of Internet of Things (IoT) technology and machine approach enables the determination of the availability of soil minerals
learning (ML) algorithms has opened up new possibilities for addressing and the corresponding prescription of fertiliser requirements [38].
these challenges. This literature review explores the existing research Fertilizer recommendation plays a crucial role in optimizing crop
and developments related to ML-enabled IoT devices for soil nutrients productivity and ensuring sustainable agricultural practices. Traditional
monitoring and crop recommendation. approaches to fertilizer recommendation often rely on expert knowledge
IoT-based soil monitoring systems have gained considerable atten­ and manual analysis, which can be time-consuming and prone to errors.
tion in recent years. These systems employ various sensors, such as However, with the advancements in machine learning techniques, there
moisture sensors, pH sensors, and nutrient sensors, to collect real-time has been growing interest in leveraging these methods [12,21,22,30] to
data on soil conditions [3,5,13,14,35,48]. The collected data is trans­ develop more accurate and efficient fertilizer recommendation systems
mitted to a central hub for analysis and decision-making. For example, [23]. represented a Machine Learning Approach to Recommend Suitable
an IoT-based system that utilized soil moisture and nutrient sensors to Crops and Fertilizers for Agriculture. They used different machine
monitor soil conditions in real-time [7,42]. The data collected was then learning approaches to perform this task and urged that random forest
used to optimize irrigation schedules and fertilizer applications, was the best performer.

2
M.R. Islam et al. Journal of Agriculture and Food Research 14 (2023) 100880

The literature review highlights the significant progress made in the programming and powering the device. It is based on the ESP8266 Wi-Fi
field of ML-enabled IoT devices for soil nutrients monitoring and crop module, which provides seamless connectivity and enables wireless
recommendation. The integration of IoT technology and ML algorithms communication with other devices and the internet. The NodeMCU of­
has proven to be effective in providing real-time data on soil conditions fers a range of functionalities and features, including 11 GPIO (General
and generating customized recommendations for farmers. These ad­ Purpose Input/Output) pins, analog-to-digital converters, and pro­
vancements hold great promise in enhancing crop productivity, opti­ grammable interfaces, making it suitable for various applications in the
mizing resource utilization, and promoting sustainable agricultural field of IoT. Its compact size, low power consumption, and compatibility
practices. Future research should focus on improving the accuracy and with the Arduino development platform, Arduino IDE, make it a popular
scalability of ML algorithms, food security at the consumer level, as well choice for IoT projects. In this research, the NodeMCU serves as a vital
as integrating additional data sources to further enhance the perfor­ component for collecting sensor data, transmitting it wirelessly using
mance of these innovative works. In the following, we provide a tabular MQTT protocol, and facilitating seamless integration with machine
summary of the most recent study in Table 1 to provide a better over­ learning algorithms.
view and distinguish our work from theirs.
3.1.2. JXBS-3001 sensor
3. Methods and materials The JXBS-3001 sensor plays a pivotal role in this research as a key
component for data collection and analysis. This sensor offers advanced
In this section, we will outline the methodologies and materials capabilities for monitoring and measuring essential soil parameters.
employed in this study, encompassing the IoT devices, sensors 3.1, data Specifically designed for agricultural applications, the JXBS-3001 sensor
transmission and analysis techniques 3.2, machine learning model enables accurate and real-time measurements of soil NPK. These mea­
design and implementation 3.3, and the proposed framework 3.4. The surements are crucial for assessing soil health, nutrient levels, and
hardware used in this research was acquired from [Link], a overall environmental conditions. The JXBS-3001 sensor boasts a robust
company based in Dhaka, Bangladesh. Different components were construction, high sensitivity, and reliable performance, making it a
sourced from various brands. The NodeMCU, produced by Espressif valuable tool for gathering precise and actionable data. It features quick
Systems in Shanghai, China, was utilized as the IoT device. The JXBS- responses, excellent interchange capabilities, and high precision agri­
3001 sensor, manufactured by Weihai JXCT Electronic Technology culture sensors. This sensor was chosen because it accurately sense data.
Company Ltd in Weihai City, China, was employed for data collection Its integration into the research framework facilitates comprehensive
purposes. Guangzhou Aosong Electronic Company Ltd., China manu­ soil monitoring, aiding in the development of innovative machine
factures the DHT11 sensor and FC-28 soil sensor is manufactured by learning algorithms for soil nutrient analysis and crop recommendation.
various suppliers.
3.1.3. FC-28 sensor
3.1. IoT devices and sensors The FC-28 sensor is a key component utilized in this research for soil
moisture monitoring. The FC-28 sensor, based on LM393 design, is
This study used three categories sensors like JXBS-3001 3.1.2, FC-28 widely recognized for its effectiveness in measuring soil moisture con­
3.1.3, DHT11 3.1.4, and one NodeMCU 3.1.1. The circuit diagram is tent. It has 4 pins for performing its operation. The sensor features two
presented in Fig. 1. And Table 2 represents the configurations and probes that are inserted into the soil, allowing it to determine the
specifications of the proposed IoT System Components. moisture level based on the electrical resistance between the probes.
With its simple design and ease of use, the FC-28 sensor provides a cost-
3.1.1. NodeMCU effective solution for monitoring soil moisture in agricultural and
The NodeMCU is an integral component used in this research for the research applications. Its analog output can be easily interfaced with
implementation of IoT capabilities. It is a versatile and widely adopted microcontrollers, enabling seamless integration into IoT systems. It has
microcontroller board. The NodeMCU board has a voltage regulator to digital output and power Indicator LEDs. With a compact PCB size of 3.2
supply a steady 3.3 V power supply as well as a USB interface for cm × 1.4 cm, the fire flame sensor offers a compact form factor for easy

Table 1
Summary of the most recent studies.
Study Purpose/Application Area Method Dataset ML-IoT Key Findings Limitations
Enabled

[35] Continuous monitoring system of soil IoT Real-time No Development and testing of a low- No Data Analysis technique was
health. data cost (5667 USD) monitoring system. conducted.
[13] Continuously measures variables at IoT and DSS Real-time No Development and testing of IoT- It must be connected with wires.
the agro-field. data. based smartphones platform.
[48] An automated remote field LoRaWAN Real-time No Visualization in the cloud. Data Analysis technique is not
monitoring system data. mentioned.
[42] Soil monitoring system IoT Real-time No Efficiently identify soil type and Data Analysis technique is not
data. display corresponding data mentioned.
graphically.
[8] Soil Classification based on ELM Private Yes Accuracy 94 % The dataset consists exclusively of the
micronutrients. dataset. Tamil Nadu region.
[6] Crop recommendation MLP Kaggle Yes 98.22 % accuracy Detailed analysis was not conducted.
Dataset
[46] To assist the farmer by developing RF Kaggle No Random forest achieved 97.18 % Not applicable for real-time data.
crop recommendation platform Dataset accuracy
[40] To support farmers with ongoing MSVM- Own dataset Yes The proposed mode achieved the Didn’t develop any platform from where
crop and field information regularly. DAG-FFO accuracy of 97.3 % farmers can get support.
[45] Crop Recommendation system RFC Own dataset Yes RFC achieved 99 % accuracy No detailed discussion about data
sensing.
[18] Correct Selection of Crop SCS Dataset from Yes Accuracy 97.4 % Dataset contains only two types of soil.
Pakistan. And the technique is only suitable for a
few crops.

3
M.R. Islam et al. Journal of Agriculture and Food Research 14 (2023) 100880

Fig. 1. Circuit diagram of the proposed system.

protocol for efficient data transmission in the ML-enabled IoT device for
Table 2
soil nutrients monitoring and crop recommendation. This section out­
Configurations and Specifications of the proposed IoT System Components.
lines the methods and techniques used for utilizing the MQTT protocol
Components Specifications for data transmission, and storage. A suitable MQTT broker needs to be
NodeMCU ESP8266 Wi-Fi module-based development board with open-source chosen, considering factors such as scalability, reliability, and support
firmware. It includes a USB interface, a for Quality of Service (QoS) levels. In this project, a cloud-based MQTT
voltage regulator for reliable power, 11 digital I/O pins, a pin for broker is chosen, as it provides easy scalability, high availability, and
analog input, and a UART communication
interface. Able to run on the Arduino IDE.
seamless integration with other cloud services. After that, MQTT topics
JXBS-3001 Designed specifically for soil monitoring applications. Its are designed to represent different aspects of soil nutrients and crop-
measurement range of NPK is 0–199 mg/kg. It uses related data, enabling efficient organization and routing of messages.
Sensor RS485 communication port and requires 5V-24v DC power supply. The collected data is formatted into MQTT payloads, encapsulating the
FC-28 Sensor The sensor operates within a voltage range of 3.3V–5V DC and has
relevant soil parameters, crop information, and other contextual data.
an operating current of 15 mA, ensuring
low power consumption during operation. LM393 based design. Based on the criticality and reliability requirements of the data, the
DHT11 The sensor provides accurate humidity readings within the range of appropriate QoS level is selected. QoS levels 0 (at most once), 1 (at least
Sensor 20 %–80 % with a precision of 5 %. It is once), or 2 (exactly once) are considered for message delivery. The IoT
also capable of accurately measuring temperatures within the range device is equipped with an MQTT client that connects to the MQTT
of 0 ◦ C–50 ◦ C with an accuracy of ±2 ◦ C.
The sampling rate of the sensor is limited to a maximum of 1 Hz.
broker and publishes the collected data to the respective MQTT topics. It
publishes data to specific topics, while other devices or subscribers can
subscribe to these topics to receive the data in real-time. For the purpose
installation and integration into fire detection systems. Therefore, the of the secured data storage and management, the MQTT broker is in­
FC-28 sensor enables researchers to obtain real-time data on soil mois­ tegrated with cloud storage services, enabling the seamless storage and
ture, facilitating informed decision-making and efficient water man­ retrieval of MQTT data in scalable storage systems. And, the received
agement strategies. MQTT messages are persisted in cloud storage for long-term data
retention and subsequent analysis. Finally, raw data received from the
3.1.4. DHT11 sensor MQTT broker undergoes preprocessing steps, including filtering and
The DHT11 sensor is an essential component used in this research for cleansing, to remove noise, outliers, or erroneous readings and is
collecting environmental data. The DHT11 sensor is widely employed transformed into a suitable format for analysis, ensuring compatibility
for measuring temperature and humidity levels. It offers a cost-effective with ML algorithms and further processing steps.
solution with reliable performance. The sensor utilizes a digital signal
output, making it compatible with various microcontrollers and IoT 3.3. Machine learning methods
platforms. With its compact size and easy integration, the DHT11 sensor
is ideal for applications where monitoring temperature and humidity are This section explains the machine learning algorithms integrated
critical. Its accuracy and stability ensure precise measurements, into the analysis pipeline to process the data and generate relevant in­
enabling researchers to gather valuable data for analysis and decision- sights and recommendations for soil nutrients monitoring and crop
making. The DHT11 sensor provides an affordable and efficient solu­ management. In our study, we used Catboost classifier 3.3.1 and
tion for environmental monitoring in research studies. Random Forest classifier with grid SearchCV 3.3.2 to recommend crop,
and fertilizer, respectively.
3.2. Data transmission using MQTT protocol
3.3.1. CatBoost Classifier
The data transmission section of the research paper focuses on the The CatBoost classifier is a powerful machine learning algorithm
implementation of the MQTT (Message Queuing Telemetry Transport) employed in the research paper. It is an open-source gradient boosting

4
M.R. Islam et al. Journal of Agriculture and Food Research 14 (2023) 100880

framework that excels in handling categorical features and large data­ find the best configuration for our model. Through rigorous evaluation
sets. In the context of the crop recommendation system, CatBoost plays a and comparison with other models, we demonstrate the effectiveness
pivotal role in predicting the most suitable crops for a given set of and efficiency of the Random Forest algorithm with grid search in
environmental and soil conditions. Its ability to handle categorical accurately predicting the optimal fertilizer recommendation for
variables without the need for extensive preprocessing makes it a different crops and soil conditions. The results highlight the potential of
preferred choice for agricultural applications. The algorithm effectively this approach in supporting precision agriculture and enhancing crop
learns from complex patterns and interactions in the data, enabling yield and sustainability. With its robustness and ability, it is an essential
accurate crop predictions. It addresses gradient bias and prediction shift component of the research paper’s fertilizer recommendation system,
issues, integrates a gradient boosting decision tree (GBDT) with cate­ providing valuable guidance to farmers and helping them make
gorical features, and concentrates on categorical variables [31]. By informed decisions to optimize crop productivity while minimizing the
using all sample datasets in the algorithm for training, it helps to in­ environmental impact.
crease the resilience of the algorithms. Assume that we observe a data
set of samples (1) 3.4. Proposed framework
S = (Xi , yi )i=1,…,n, (1)
The proposed framework involves the installation of three sen­
Where Xi = (x1i , x2i , …, xki ) is a vector of k characteristics and response sors—the JXBS-3001, FC-28, DHT11 sensor—in the crop field to collect
feature yiϵF, which can be encoded as the numerical feature (0,1,2,3, …., various types of data, including soil nutrient concentrations (NPK),
m). The distribution of the samples (Xi, yi) is independently and iden­ moisture, humidity, and temperature. The collected data is then trans­
tically distributed according to an unidentified distribution, d(., .). The mitted to a remote server through the NodeMCU using the MQTT pro­
learning task’s objective is to develop a function A: Fk → F that mini­ tocol for further processing. These sensors provide real-time data,
mizes the estimated loss indicated in equation (2) allowing farmers to monitor soil health and make informed decisions
regarding fertilization strategies and irrigation schedules. This data is
Los(A) := EL(y, A(X)) (2) passed to the proposed machine learning model, which generates
where L(., .) is a smooth loss function and (X, y) is a sample of test personalized crop recommendations, including ideal planting times,
data drawn from the training data S [17]. Additionally, CatBoost offers suitable crop varieties, and precise fertilization plans. This empowers
features like robust handling of missing values, strong regularization farmers to optimize their crop selection and improve overall produc­
techniques, and customizable parameters that can be fine-tuned for tivity. Besides, it ensures the health and safety of harvested produce by
optimal performance. By leveraging the capabilities of CatBoost, the soil monitoring and assesses nutritional content, pesticide residues, and
research paper’s crop recommendation system aims to provide farmers potential contaminants in the food by analyzing data from embedded
and agricultural stakeholders with valuable insights and informed sensors. The proposed framework offers a User-Friendly Mobile App,
decision-making tools to enhance crop productivity and sustainability. which can provides farmers with seamless access to their device and
data, allowing them to remotely monitor and control their agricultural
3.3.2. Random forest with grid SearchCV operations. Users can view real-time sensor readings, track historical
Random Forest is an ensemble learning method that combines mul­ trends, receive actionable insights, and access crop recommendations
tiple decision trees to make predictions. In the context of the fertilizer and food health reports, all within the app’s intuitive interface. It also
recommendation, Random Forest plays a crucial role in analyzing delivers real-time alerts and notifications to farmers regarding critical
various factors such as soil properties, crop characteristics, and envi­ conditions such as water stress, nutrient deficiencies, disease outbreaks,
ronmental conditions to suggest the most appropriate fertilizer for or adverse weather conditions, and presents comprehensive data visu­
optimal plant growth. The algorithm’s ability to handle complex in­ alization and analysis tools. Farmers can view sensor data in intuitive
teractions and nonlinear relationships between variables makes it well- graphs and charts, identify patterns, and gain valuable insights into soil
suited for this task. By utilizing an ensemble of decision trees, Random health, crop performance, and food quality. Overall, the proposed
Forest reduces overfitting and improves prediction accuracy. It also framework utilizes a combination of IoT sensors and machine learning
provides insights into feature importance, aiding in understanding the algorithms to enable remote agro-field monitoring and crop recom­
factors that influence fertilizer recommendations. The base classifiers mendation. By leveraging the power of machine learning, the system can
can be defined as (3) accurately analyze vast amounts of data and makes informed decisions
about agro-field conditions, leading to better outcomes, food safety, and
{t(x, φm ), m = 1, …….., } (3) improved food production. The proposed framework is illustrated in
Where, x is input vector and {φm} are the independent and identi­ Fig. 2.
cally distributed random vectors [39]. Then, the random forest approach
can be defined as (4) 4. Experimental results

Rf = majorityVote{t(x, φm ), m = 1, …….., } (4) In this section, we will provide an explanation of the outcomes
achieved in our study. This will encompass field deployment design 4.1
In this research paper, we explore the application of the Random Forest
and information about the techniques employed for data collection and
algorithm in combination with grid search for optimizing fertilizer processing 4.2, along with a summary of the accuracy measures 4.3
recommendation systems. Random Forest is a popular ensemble employed to assess the effectiveness of our system. Lastly, we will
learning method that combines multiple decision trees to make accurate conduct an analysis of the acquired results 4.4 and introduce an IoT-ML
predictions. Grid search, on the other hand, is a hyperparameter tuning enabled agriculture platform 4.5.
technique that systematically searches through a specified parameter
grid to find the optimal combination of hyperparameters for a machine
4.1. Experimental setup
learning model. By utilizing grid search, we aim to fine-tune the Random
Forest model to improve its performance in fertilizer recommendation.
The study was conducted using an 8th generation Intel Corei7 with a
We conduct experiments on a comprehensive dataset containing soil
6600U processor clocked at up to 3.1 GHz and 16 GB of RAM. Numerous
properties, climate data, and crop information. The grid search tech­
tests were conducted using various machine learning techniques using
nique allows us to explore various hyperparameter combinations, such
the Anaconda tool. To gather diverse data from the crop field, three
as the number of trees, maximum depth, and minimum sample split, to
sensors—the JXBS-3001, FC-28, and DHT11 sensors—are deployed.

5
M.R. Islam et al. Journal of Agriculture and Food Research 14 (2023) 100880

Fig. 2. Proposed framework of an innovative ML-Enabled IoT system for soil nutrients monitoring and crop recommendation.

Deploying sensors and controllers in an IoT-based agricultural system is 4.2. Data collection and pre-processing
crucial for precise data collection. The key considerations of the
deployment in field include strategic sensor placement (at various To initiate the learning and validation process of the proposed
depths in the soil) to monitor soil properties at different levels, even framework, we initially utilized a combined dataset compiled from
distribution for accurate data capture, node controllers (NodeMCU) near Ref. [19] and our own collected dataset from diverse regions of
sensors, power supply via solar panels for remote areas, reliable Bangladesh for crop recommendation which contains 2299 records. A
communication networks (LoRaWAN), weatherproofing for durability, diverse set of 22 crop types data, including apple, banana, black gram,
regular calibration and maintenance, centralized data storage, user- grapes, kidney bean, chickpea, coconut, coffee, cotton, jute, lentil,
friendly interfaces, and scalability. These sensors enable the collection maize, moth beans, mung beans, pomegranate, pigeon peas, musk­
of information such as soil nutrient concentrations (NPK), moisture melon, orange, papaya, rice, and watermelon, was collected and utilized
levels, humidity, and temperature. The acquired data is subsequently for the development of our crop recommendation system. This
transmitted to a remote cloud via the NodeMCU device, utilizing the comprehensive dataset represents a wide spectrum of crops commonly
MQTT protocol. Fig. 3 illustrates the field experimental design of the cultivated in various agricultural regions, ensuring the system’s appli­
ML-Enabled IoT system for monitoring soil nutrients and crop cability across different farming contexts. This combined dataset allows
recommendations. us to recommend a suitable crop for the agri-field based on its 7 features,
such as Nitrogen (N), Potassium (K), Phosphorus (P), Temperature,
Humidity, and soil Moisture. Additionally, we employed a distinct

Fig. 3. Field Experimental Design of the ML-Enabled IoT System for monitoring soil nutrients and crop recommendations.

6
M.R. Islam et al. Journal of Agriculture and Food Research 14 (2023) 100880

dataset to learn and validate our proposed system specifically for fer­
tilizer recommendation based on its 9 features, such as Nitrogen (N),
Potassium (K), Phosphorus (P), Temperature, Humidity, Soil type, Crop
type, and soil Moisture. This dataset contains 152 entries prepared from
our own data and a Kaggle dataset [43]. By using the chi-squared test
(5), we can evaluate the independence or dependence between features,
identify patterns or relationships, and make informed conclusions about
the significance of associations within the data.
∑ (Obi − Exi )
Ci2 = (5)
Exi

Where, Obi is the observed value, and Exi is the expected value. Table 3
represents the chi-squared score of the crop recommendation dataset,
where the score of ph is worst compared with other features. Therefore,
we eliminate the ph from the dataset for further processing. Then, we
apply the Z-score (6) to remove outliers from the dataset.
(x − μ)
Z= (6)
σ
Where, x is value, μ is mean, and σ is standard deviation. We have Fig. 4. Correlation among the features of the crop recommendation dataset.
established a threshold to identify and exclude outliers from the dataset.
Any data point with a z-score less than 3 is considered an outlier and 4.3.2. Precision
removed. Through this process, we detected 9 outliers, resulting in a Precision 9 is determined by dividing the number of true positive
revised dataset size of 2299 instances, compared to the original 2290 samples by the total number of positive samples predicted by the model:
instances. Fig. 4 shows the correlation among the features of the crop
recommendation dataset. Precision =
(TP)
(9)
We convert the categorical features of the fertilizer prediction (TP + FP)
dataset into numerical values. Then, we performed Z-score methods to
remove outliers. We did not find any outliers from the dataset. To 4.3.3. Recall
convert the ppm unit of soil nutrients into Kg/HA the following equation Recall (10), also known as the true positive rate (TPR), is calculated
(7) is used [44]. by dividing the number of correctly recommended or predicted crops or
fertilizers by the total number of crops or fertilizers. It can be defined as:
Nutrient(kg / HA) = 2.5 ∗ λ (7)
(TP)
Recall = (10)
Where, λ is PPM(parts per million) of Nutrient. (TP + FN)
After preprocessing, the data set was split into a training set (80 %)
and a test set (20 %). Although the model was primarily trained and 4.3.4. F1-measure
assessed using the combined Dataset, we also conducted tests using real- It is the precision and recall’s harmonic mean (average) defined in
time data collected in our laboratory. equation (11).

4.3. Model evaluation metrics F1 − Score =

(2 ∗ Precision ∗ Recall)
(11)
(Precision + Recall)
Prior to implementing our model on simulated data, it is essential to
enhance its overall predictive capability by assessing its performance
4.4. Results analysis
using diverse criteria. It is important to evaluate an ML model using
various metrics [29] and not solely relying on accuracy when applying it
The results obtained from the implementation of the CatBoost clas­
to new data, as this can result in inaccurate predictions. To evaluate the
sifier in the crop recommendation 4.4.1 and from the implementation of
effectiveness of our research, we employ the following evaluation
the Random Forest classifier with Grid Search in the fertilizer prediction
criteria (Accuracy 4.3.1, Precision 4.3.2, Recall 4.3.3, F1-Measur 4.3.4)
4.4.2 are analyzed in this subsection. The performance of the classifiers
with TP- True Positive, TN- True Negative, FP- False Positive, and FN-
was evaluated using various evaluation metrics, including accuracy,
False Negative.
precision, recall, and F1-score.

4.3.1. Accuracy
4.4.1. Crop recommendation
Accuracy refers (8) to the proportion of accurately classified samples
The accuracy metric indicates the overall correctness of the crop
compared to the total number of samples within the dataset:
recommendations provided by the system. Precision represents the
(TP + TN) proportion of correctly recommended crops out of all the crops sug­
Accuracy = (8)
(TP + TN + FP + FN) gested by the system, while recall measures the system’s ability to
identify and recommend the relevant crops. The F1-score combines both
precision and recall to provide a balanced assessment of the classifier’s
performance.

Table 3
Chi-squared score of the crop recommendation dataset.
N P K temperature humidity ph rainfall

53144.69 42500.13 116710.53 1092.42 14755.48 74.88 54808.13

7
M.R. Islam et al. Journal of Agriculture and Food Research 14 (2023) 100880

Fig. 5 represents the comparisons of 3 different model evaluation

metrics which indicate the effectiveness of the applied CatBoost Clas­
sifier with Voting and Bagging classifiers in terms of accuracy, precision,
Recall, and F1- measure. The CatBoost classifier achieved the best per­
formance with an accuracy of 97.5 %, precision of 98 %, recall of 96 %,
and F1-score of 97.5 %. Other models also perform well. The bagging
classifier obtained the accuracy of 95 %, precision of 96 %, recall of 95
%, and F1-score of 95 %, while the voting classifier acquired the accu­
racy of 96 %, precision of 94 %, recall of 96 %, and F1-score of 97 %.
Additionally, the confusion matrix was analyzed to gain insights into
the specific types of classification errors made by the system presented
in Fig. 6. Upon analyzing the results, it is evident that the model exhibits
a significant number of accurate positive and negative predictions across
various classes. Nonetheless, there are instances where the model gen­
erates incorrect outputs. Specifically, there are 52 occurrences where the
model misclassifies certain classes. These errors may have the potential
to impact crop selection decisions within the agricultural domain.
Table 4 represents a comparative performance analysis of different
machine learning algorithms used in the recent literature to recommend
the suitable crop for the cultivation in the agri-field. It indicates that the
CatBoost classifier demonstrates promising performance with 97.5 % Fig. 6. Confusion matrix of CatBoost classifier.
accuracy and 97 % F1 score in accurately recommending crops based on
the given input parameters. These findings validate the effectiveness of predicting the appropriate fertilizer for optimal crop cultivation.
the implemented crop recommendation system and highlight the po­ Fig. 8 represents a comparative performance analysis of different
tential of the CatBoost classifier in agricultural decision-making machine learning algorithms used in the recent literature to predict the
processes. suitable fertilizer for crop cultivation in the agri-field. It indicates that
the Random Forest with Grid SearchCV demonstrates promising per­
4.4.2. Fertilizer prediction formance with 99.56 % accuracy and 100 % F1 score in accurately
After several iterations and fine-tuning, we successfully developed a predicting fertilizers based on the given input parameters. In compari­
fertilizer prediction model using the Random Forest algorithm combined son to alternative machine learning methods like SVM, Random Forest,
with grid searchCV. The hyperparameters of the model were carefully and Linear Forest, the proposed approach exhibited higher accuracy and
adjusted to achieve optimal performance. To determine the best values F1-score, demonstrating its superior performance in fertilizer predic­
for these hyperparameters, we conducted a cross-validated grid search, tion. To illustrate, previous studies by Ref. [12] achieved an accuracy of
as depicted in Table 5 of our study. This process involved systematically 0.97 and an F1-score of 0.95 using Linear Forest, while [22] reported an
exploring different combinations of hyperparameters and evaluating accuracy of 0.95 and an F1-score of 0.95 with Random Forest. The
their performance using cross-validation techniques. The resulting proposed method surpassed these approaches, including advanced
model with the identified optimal hyperparameter values demonstrates techniques such as support vector machines. For instance Ref. [30],
promising potential for accurate fertilizer prediction in our research. achieved an accuracy of 0.95 and an F1-score of 0.894 with an SVM
The confusion matrix was analyzed to gain insights into the specific model. Overall, these findings validate the effectiveness of the imple­
types of classification errors made by the system presented in Fig. 7. mented fertilizer prediction system and highlight the potential of the
Upon analyzing the results, it is evident that the model exhibits a sig­ Random Forest with Grid SearchCV classifier in agricultural
nificant number of accurate positive and negative predictions across decision-making processes.
various classes. It is mentionable that there is only one instance where
the model generates incorrect output. 4.5. IoT-ML enabled agriculture platform
Table 6 presents the precision, recall, and f1-score for each class,
indicating the model’s performance. It is evident that the model exhibits Based on the evaluation criteria discussed earlier, it is evident that
strong performance for most classes, with a precision of 1.00 and a recall CatBoost demonstrates superior performance compared to the other
value of 0.99. This suggests that the model excels at accurately machine learning algorithms considered in predicting the optimal crop
for harvesting. Therefore, for our crop recommendation platform, we
plan to utilize CatBoost to predict the most suitable crop based on user-
input parameters such as N, P, K, air temperature, air humidity, soil pH
level, moisture, and rainfall. Similarly, it has been observed that RF with
gridsearchCV yields the best results in predicting the appropriate fer­
tilizer for field application. Hence, for our platform, we will employ RF
with gridsearchCV to recommend the most suitable fertilizer for the
earlier recommended crop, considering user-provided parameters such
as Nitrogen (N), Potassium (K), Phosphorus (P), Temperature, Humidity,
Soil type, Crop type, and Soil Moisture. Once the optimal models have
been selected, they will be saved separately for the development of the
crop recommendation platform using Flutter as the subsequent step.
Following successful local testing, we deployed our locally tested mobile
app on Google Cloud App Engine, which offers Platform as a Service
(PaaS) capabilities, ensuring high uptime of nearly 99.9 % and conve­
nient access from any device around the clock. For demonstration pur­
poses, we utilized their free tier service, which does not incur any
Fig. 5. Comparisons of 3 different model’s evaluation metrics. charges. Fig. 9 provides a glimpse of the fully functional

8
M.R. Islam et al. Journal of Agriculture and Food Research 14 (2023) 100880

Table 4
A comparative performance analysis of ML algorithms used in the recent literature to recommend the suitable crop.
Ref Year Dataset Sensors Used Methodology Performance

Bakthavatchalam et al. [30] 2022 Kaggle dataset Yes MLP Accuracy: 98 %, F1 Score:
Thilakarathne et al. [31] 2022 Kaggle dataset No Random Forest Accuracy: 97.18 %, F1 Score: 97 %
Senapaty et al. [32] 2023 Field data of India Yes MSVM-DAG-FFO Accuracy: 97.3 %, F1 Score: 96 %
Sundaresan et al. [33] 2022 Own data Yes KNN Accuracy: 97 %, F1 Score:
Ikram et al. [36] 2022 Own data Yes Ensemble approach Accuracy: 97.4 %, F1 Score: 97 %
Proposed Model 2023 Kaggle dataset + own dataset Yes CatBoost Accuracy: 97.5 %, F1 Score: 97.5 %

results obtained from our research indicate that the proposed device and
Table 5 framework have significant potential in the field of precision
The hyperparameter tuned with a grid search for the random forest.
agriculture.
Hyperparameter Grid Search Parameter One of the key contributions of our study is the integration of IoT
Number of trees 200, 500 technology with machine learning algorithms for real-time soil nutrient
Maximum features auto, sqrt, log2 monitoring. By deploying sensors such as JXBS-3001, FC-28, and DHT11
Maximum tree depth 4,5,6,7,8 in the crop field, we were able to collect essential data related to soil
Cross Validation 5
nutrient concentrations (NPK), moisture, humidity, and temperature.
This data was then transmitted to a remote server using the NodeMCU
and MQTT protocol, enabling continuous monitoring and analysis.
The use of machine learning algorithms, particularly CatBoost and
Random Forest, played a crucial role in the crop recommendation pro­
cess. These algorithms were trained on merged datasets and optimized
through hyperparameter tuning, resulting in accurate predictions for the
most suitable crops and fertilizers. The CatBoost algorithm exhibited
superior performance in predicting the optimal crop based on input
parameters such as N, P, K, air temperature, air humidity, soil pH level,
moisture, and rainfall. On the other hand, the Random Forest algorithm
with grid search excelled in recommending the most appropriate fer­
tilizer based on parameters such as Nitrogen (N), Potassium (K), Phos­
phorus (P), Temperature, Humidity, Soil type, Crop type, and Soil
Moisture.
The performance evaluation of our models revealed high precision,
recall, and F1-scores for most classes, indicating the effectiveness of our
Fig. 7. Confusion Matrix of Random Forest Classifier with gridSearchCV. approach in crop and fertilizer recommendation. However, it is impor­
tant to note that there were instances where the models made errors,
which could have implications for crop selection and fertilization de­
Table 6 cisions. Further research and refinement are necessary to minimize these
Classification Report for the proposed model performance.
errors and improve the overall accuracy of the system.
Class precision recall f1-score Support The integration of the ML-enabled IoT device with a cloud-based
10-26-26 1.00 1.00 1.00 2 recommendation platform provides convenient access and scalability.
14-35-14 1.00 1.00 1.00 4 By deploying the locally tested mobile app on Google Cloud App Engine,
17-17-17 1.00 1.00 1.00 1 we ensured high uptime and easy accessibility from any device. The
20–20 1.00 0.75 0.86 4
28–28 1.00 1.00 1.00 5
cloud-hosted platform offers a fully functional recommendation system
DAP 1.00 1.00 1.00 5 for farmers, enabling them to make informed decisions regarding crop
Urea 1.00 1.00 1.00 11 selection and fertilizer application.
accuracy 0.99‘ 32 Future research directions for the paper include.
macro avg 1.00 0.96 0.98 32
weighted avg 1.00 0.99 1.00 32
1. Expanding capabilities: The device and platform should be further
developed to accommodate a broader range of crops and geographic
recommendation platform hosted in the cloud. The platform can locations. This expansion would allow farmers cultivating different
determine the name and quantity of applied fertilizers and the list of types of crops in various regions to benefit from the device’s moni­
potential high yielding crops according to the soil fertility test and test toring and recommendation functionalities.
results. Fertilizers and treatments applied to the field during production 2. Integration of satellite imagery and weather data: To provide more
are stored in the database. As a result, it has become possible to deter­ comprehensive and precise recommendations, future research
mine the quality of the produce at the consumer level through the mo­ should focus on incorporating satellite imagery and weather data
bile app when it is marketed. The use of QR codes allows farmers to into the system. By leveraging these additional sources of informa­
market their crops effectively. Therefore, consumers have the ability to tion, the device can offer more accurate insights into crop health,
access detailed crop history information, such as the planting time, and growth patterns, and environmental conditions.
the type and quantity of fertilizers used, etc. 3. Enhancing scalability and usability: The device’s scalability and us­
ability should be improved to cater to the needs of farmers in
5. Discussion and future research different regions with varying technological infrastructures. This
includes ensuring that the device can adapt to different communi­
The aim of this study was to develop an innovative ML-enabled IoT cation networks and power sources, as well as developing user-
device for soil nutrients monitoring and crop recommendation. The

9
M.R. Islam et al. Journal of Agriculture and Food Research 14 (2023) 100880

Fig. 8. A comparative performance analysis of ML algorithms used in the recent literature to predict the suitable fertilizer.

Fig. 9. Crop recommendation platform hosted in the cloud.

friendly interfaces that are accessible to farmers with varying levels adaptability, and real-time decision support. Nonetheless, traditional
of technological literacy. methods retain their inherent merits, encompassing simplicity, reli­
ability, cost-effectiveness, transparency, and wide applicability. The
By addressing these future research directions, the paper can contribute selection between the traditional and ML-Enabled IoT approaches
to the advancement of the ML-enabled IoT device and its applicability in hinges on factors such as the precise requirements of the agricultural
agriculture, enabling farmers to make more informed decisions and context, data and resource availability, and the users’ technical exper­
improve crop productivity. tise. Notably, this system holds substantial promise in assisting the many
Overall, our research demonstrates the potential of ML-enabled IoT farmers worldwide who lack comprehensive agricultural knowledge,
devices in precision agriculture. The integration of sensor data, machine making it especially valuable for such individuals. Future research en­
learning algorithms, and cloud-based platforms enables real-time deavors should concentrate on fine-tuning the system and broadening its
monitoring, analysis, and decision-making for improved crop manage­ capabilities to cater to the diverse needs of farmers and agricultural
ment practices. This innovative system offers numerous benefits, stakeholders.
including unparalleled precision, real-time monitoring capabilities,
adaptability, user-friendliness, support for safe food production, and the 6. Conclusion
potential for sustainability. However, it also faces challenges related to
data availability, sensitivity to errors, system complexity, cost consid­ In this study, the combination of IoT technology, sensor data
erations, and ongoing maintenance requirements. The system’s suit­ collection, and machine learning algorithms has demonstrated prom­
ability is dependent on the unique demands and resources of the specific ising results in precision agriculture through an innovative ML-enabled
agricultural context in which it is deployed. Furthermore, when IoT device for soil nutrient monitoring and crop recommendation. By
comparing our system with the conventional approach, it becomes utilizing various sensors and transmitting data to a remote server in real-
evident that the proposed system excels in terms of precision, time, crucial information on soil nutrient concentrations, moisture,

10
M.R. Islam et al. Journal of Agriculture and Food Research 14 (2023) 100880

humidity, and temperature was collected and analyzed. The application [13] S.V. Gaikwad, A.D. Vibhute, K.V. Kale, S.C. Mehrotra, An innovative iot based
system for precision farming, Comput. Electron. Agric. 187 (2021), 106291.
of machine learning algorithms, such as CatBoost and Random Forest,
[14] L. García, L. Parra, J.M. Jimenez, J. Lloret, P. Lorenz, Iot-based smart irrigation
proved effective in predicting suitable crops and recommending systems: an overview on the recent trends on sensors and iot systems for irrigation
appropriate fertilizers. While there is room for improvement in mini­ in precision agriculture, Sensors 20 (4) (2020) 1042, [Link]
mizing errors and refining the models, this ML-enabled IoT device has s20041042.
[15] X. Ge, J. Wang, J. Ding, X. Cao, Z. Zhang, J. Liu, X. Li, Combining uav-based
shown great potential in providing farmers with real-time insights and hyperspectral imagery and machine learning algorithms for soil moisture content
recommendations for optimal crop management. By addressing the monitoring, PeerJ 7 (2019), e6926, [Link]
challenges and further exploring research directions, we can continue to [16] T. Hengl, J.G. Leenaars, K.D. Shepherd, M.G. Walsh, G.B. Heuvelink, T. Mamo,
H. Tilahun, E. Berkhout, M. Cooper, E. Fegraus, et al., Soil nutrient maps of sub-
enhance the capabilities of this device and contribute to the advance­ saharan africa: assessment of soil nutrient content at 250 m spatial resolution using
ment of precision agriculture. machine learning, Nutrient Cycl. Agroecosyst. 109 (2017) 77–102.
[17] A.A. Ibrahim, R.L. Ridwan, M.M. Muhammed, R.O. Abdulaziz, G.A. Saheed,
Comparison of the catboost classifier with other machine learning methods, Int. J.
CRediT authorship contribution statement Adv. Comput. Sci. Appl. 11 (11) (2020).
[18] A. Ikram, W. Aslam, R.H.H. Aziz, F. Noor, G.A. Mallah, S. Ikram, M.S. Ahmad, A.
M. Abdullah, I. Ullah, Crop yield maximization using an iot-based smart decision,
Md. Reazul Islam: Conceptualization of this study, Methodology,
J. Sens. (2022), [Link]
Software, Data curation. Khondokar Oliullah: Conceptualization of this [19] A. Ingle, Crop recommendation dataset. [Link]
study, Data curation, Writing - Original draft preparation, Methodology, rvaingle/crop-recommendation-dataset, 2020.
Investigation. Md Mohsin Kabir: Writing - Original draft preparation, [20] M.R. Islam, M.M. Kabir, M.F. Mridha, S. Alfarhood, M. Safran, D. Che, Deep
learning-based iot system for remote monitoring and early detection of health
Investigation, Supervision, Visualization, Writing- Review & Editing. issues in real-time, Sensors 23 (11) (2023) 5204, [Link]
Munzirul Alom: Data curation, Methodology, Investigation. M. F. s23115204.
Mridha: Investigation, Supervision, Visualization, Writing- Review & [21] D. Jayashree, O. Pandithurai, L. Paul Jasmin Rani, P.K. Menon, M.V. Beria,
S. Nithyalakshmi, Fertilizer recommendation system using machine learning, in:
editing. Disruptive Technologies for Big Data and Cloud Applications: Proceedings of
ICBDCC 2021, Springer, 2022, pp. 709–716.
[22] R. Jeevaganesh, D. Harish, B. Priya, A machine learning-based approach for crop
Declaration of competing interest yield prediction and fertilizer recommendation, in: 2022 6th International
Conference on Trends in Electronics and Informatics (ICOEI), IEEE, 2022,
pp. 1330–1334.
The authors certify that they have NO affiliations with or involve­ [23] G.K. Jha, P. Ranjan, M. Gaur, A machine learning approach to recommend suitable
ment in any organization or entity with any financial interest or non- crops and fertilizers for agriculture. Recommender System with Machine Learning and
Artificial Intelligence: practical Tools and Applications in Medical, Agricultural and
financial interest (such as personal or professional relationships, affili­
Other Industries (2020) 89–99, [Link]
ations, knowledge or beliefs) in the subject matter or materials discussed [24] F. Kiani, A. Seyyedabbasi, S. Nematzadeh, F. Candan, T. Çevik, F.A. Anka,
in this manuscript. G. Randazzo, S. Lanza, A. Muzirafuti, Adaptive metaheuristic-based methods for
autonomous robot path planning: sustainable agricultural applications, Appl. Sci.
12 (3) (2022) 943, [Link]
Data availability [25] W. Li, T. Logenthiran, V.-T. Phan, W.L. Woo, A novel smart energy theft system
(sets) for iot-based smart home, IEEE Internet Things J. 6 (3) (2019) 5531–5539.
Data will be made available on request. [26] M. Mulge, M. Sharnappa, A. Sultanpure, D. Sajjan, M. Kamani, Agricultural crop
recommendation system using iot and ml, Int. J. Anal. Exp. Modal Anal. 12 (6)
(2020).
References [27] K. Nischitha, D. Vishwakarma, M.N. Ashwini, M. Manjuraju, Crop prediction using
machine learning approaches, Int. J. Eng. Res. Technol. 9 (8) (2020) 23–26.
[28] K. Oliullah, S. Tarafdar, M. Whaiduzzaman, Fog computing and iot based modern
[1] Y. Al Masmoudi, Y. Bouslihim, K. Doumali, L. Hssaini, K. Ibno Namr, Use of
agricultural system, Topics in Intell Comput Ind Design 2 (1) (2020) 88–92,
machine learning in moroccan soil fertility prediction as an alternative to laborious
[Link]
analyses, Modeling Earth Systems and Environment (2021) 1–11, [Link]
[29] K. Oliullah, A. Barros, M. Whaiduzzaman, Analyzing the effectiveness of several
10.1007/s40808-021-01329-8.
machine learning methods for heart attack prediction, in: Proceedings of the
[2] J. An, F. Le Gall, J. Kim, J. Yun, J. Hwang, M. Bauer, M. Zhao, J. Song, Toward
Fourth International Conference on Trends in Computational and Cognitive
global iot-enabled smart cities interworking using adaptive semantic adapter, IEEE
Engineering: TCCE 2022, Springer, 2023, pp. 225–236.
Internet Things J. 6 (3) (2019) 5753–5765, [Link]
[30] A. Palaniraj, A. Balamurugan, R.D. Prasad, P. Pradeep, Crop and fertilizer
JIOT.2019.2905275.
recommendation system using machine learning, IRJET 8 (2021).
[3] N. Ananthi, J. Divya, M. Divya, V. Janani, Iot based smart soil monitoring system
[31] L. Prokhorenkova, G. Gusev, A. Vorobev, A.V. Dorogush, A. Gulin, Catboost:
for agricultural production, in: 2017 IEEE Technological Innovations in ICT for
unbiased boosting with categorical features, Adv. Neural Inf. Process. Syst. (2018)
Agriculture and Rural Development (TIAR), IEEE, 2017, pp. 209–214.
31.
[4] A. Ashwitha, C. Latha, Crop recommendation and yield estimation using machine
[32] Z. Qiu, F. Ma, Z. Li, X. Xu, H. Ge, C. Du, Estimation of nitrogen nutrition index in
learning, Journal of Mobile Multimedia (2022) 861–884, [Link]
rice from uav rgb images coupled with machine learning algorithms, Comput.
10.13052/jmm1550-4646.18320.
Electron. Agric. 189 (2021), 106421, [Link]
[5] G. Babu Loganathan, E. Mohan, R. Siva Kumar, Iot based water and soil quality
compag.2021.106421.
monitoring system, Int. J. Mech. Eng. Technol. 10 (2) (2019) 537–541.
[33] V.K. Quy, N.V. Hau, D.V. Anh, N.M. Quy, N.T. Ban, S. Lanza, G. Randazzo,
[6] K. Bakthavatchalam, B. Karthik, V. Thiruvengadam, S. Muthal, D. Jose, K. Kotecha,
A. Muzirafuti, Iot-enabled smart agriculture: architecture, applications, and
V. Varadarajan, Iot framework for measurement and precision agriculture:
challenges, Appl. Sci. 12 (7) (2022) 3396, [Link]
predicting the crop using machine learning algorithms, Technologies 10 (1) (2022)
[34] V.K. Quy, V.H. Nam, D.M. Linh, N.T. Ban, N.D. Han, Communication solutions for
13.
vehicle ad-hoc network in smart cities environment: a comprehensive survey,
[7] V. Bhatnagar, R. Chandra, Iot-based soil health monitoring and recommendation
Wireless Pers. Commun. (2022) 1–25.
system, Internet of Things and Analytics for Agriculture 2 (2020) 1–21, [Link]
[35] S.J. Ramson, W.D. León-Salas, Z. Brecheisen, E.J. Foster, C.T. Johnston, D.
org/10.1007/978-981-15-0663-5_1.
G. Schulze, T. Filley, R. Rahimi, M.J.C.V. Soto, J.A.L. Bolivar, et al., A self-
[8] T. Blesslin Sheeba, L. Anand, G. Manohar, S. Selvan, C.B. Wilfred, K. Muthukumar,
powered, real-time, lorawan iot-based soil health monitoring system, IEEE Internet
S. Padmavathy, P. Ramesh Kumar, B.T. Asfaw, Machine learning algorithm for soil
Things J. 8 (11) (2021) 9278–9293.
analysis and classification of micronutrients in iot-enabled automated farms,
[36] S. Rani, S.H. Ahmed, S.C. Shah, Smart health, A novel paradigm to control the
J. Nanomater. (2022), [Link]
chickungunya virus, IEEE Internet Things J. 6 (2) (2018) 1306–1311, [Link]
[9] D.A. Bondre, S. Mahagaonkar, Prediction of crop yield and fertilizer
org/10.1109/JIOT.2018.2802898.
recommendation using machine learning algorithms, International Journal of
[37] D.A. Reddy, B. Dadore, A. Watekar, Crop recommendation system to maximize
Engineering Applied Sciences and Technology 4 (5) (2019) 371–376.
crop yield in ramtek region using machine learning, International Journal of
[10] A. Chehri, H. Chaibi, R. Saadane, N. Hakem, M. Wahbi, A framework of optimizing
Scientific Research in Science and Technology 6 (1) (2019) 485–489, [Link]
the deployment of iot for precision agriculture industry, Procedia Comput. Sci. 176
org/10.32628/IJSRST196172.
(2020) 2414–2422.
[38] R. Reshma, V. Sathiyavathi, T. Sindhu, K. Selvakumar, L. SaiRamesh, Iot based
[11] F. Cirillo, D. Gómez, L. Diez, I.E. Maestro, T.B.J. Gilbert, R. Akhavan, Smart city iot
classification techniques for soil content analysis and crop yield prediction, in:
services creation through large-scale collaboration, IEEE Internet Things J. 7 (6)
2020 Fourth International Conference on I-SMAC (IoT in Social, Mobile, Analytics
(2020) 5267–5275, [Link]
and Cloud)(I-SMAC), IEEE, 2020, pp. 156–160.
[12] K. V. P. Dr Kalli Srinivasa Nageswara Prasad, S.V.V.D. Venu Gopal, Suitable
fertilizer recommendation system using linear forest classifier, Ymer 21 (2022).

11
M.R. Islam et al. Journal of Agriculture and Food Research 14 (2023) 100880

[39] V.F. Rodriguez-Galiano, B. Ghimire, J. Rogan, M. Chica-Olmo, J.P. Rigol-Sanchez, [46] N.N. Thilakarathne, M.S.A. Bakar, P.E. Abas, H. Yassin, A cloud enabled crop
An assessment of the effectiveness of a random forest classifier for land-cover recommendation platform for machine learning-driven precision farming, Sensors
classification, ISPRS J. Photogrammetry Remote Sens. 67 (2012) 93–104, https:// 22 (16) (2022) 6299.
[Link]/10.1016/[Link].2011.11.002. [47] A. Vangala, A.K. Das, V. Chamola, V. Korotaev, J.J. Rodrigues, Security in iot-
[40] M.K. Senapaty, A. Ray, N. Padhy, Iot-enabled soil nutrient analysis and crop enabled smart agriculture: architecture, security solutions and challenges, Cluster
recommendation model for precision agriculture, Computers 12 (3) (2023) 61. Comput. 26 (2) (2023) 879–902, [Link]
[41] D. Shin, K. Yun, J. Kim, P.V. Astillo, J.-N. Kim, I. You, A security protocol for route [48] B.M. Zerihun, T.O. Olwal, M.R. Hassen, Design and analysis of iot-based modern
optimization in dmm-based smart home iot networks, IEEE Access 7 (2019) agriculture monitoring system for real-time data collection, in: Computer Vision
142531–142550, [Link] and Machine Learning in Agriculture, Springer, 2022, pp. 73–82, [Link]
[42] V. Shukla, S. Dixit, P. Dixit, An iot based user authenticated soil monitoring system, 10.1007/978-981-16-9991-7_5, 2.
Ad Hoc Sens. Wirel. Netw. 53 (2022). [49] X. Zhang, J. Zhang, L. Li, Y. Zhang, G. Yang, Monitoring citrus soil moisture and
[43] M. Singh, Fertilizer Prediction Dataset, 2022. [Link] nutrients using an iot based system, Sensors 17 (3) (2017) 447.
mohitsingh1804/fertilizer-prediction-dataset. [50] Z. Zhou, H. Yu, H. Shi, Human activity recognition based on improved bayesian
[44] K.S.R. Subramanian, Design and implementation of fertilizer recommendation convolution network to analyze health care data using wearable iot device, IEEE
system for farmers, Test Eng. Manag. 83 (2020) 8840–8849. Access 8 (2020) 86411–86418, [Link]
[45] S. Sundaresan, S.D. Johnson, V.M. Bharathy, P.M.P. Kumar, M. Surendar, Machine
learning and iot-based smart farming for enhancing the crop yield, in: Journal of
Physics: Conference Series vol. 2466, IOP Publishing, 2023, 012028.

12