2024 IEEE International Conference on Interdisciplinary Approaches in Technology and Management for Social Innovation (IATMSI) | 979-8-3503-6052-3/24/$31.

00 ©2024 IEEE | DOI: 10.1109/IATMSI60426.2024.10502731

IoT Based Precise Greenhouse Management System

using Machine Learning Algorithm
Ashay Rokade Manwinder Singh
Department of Electronics & Telecommunication Engg. School of Electronics and Electrical Engineering
Prof. Ram Meghe Institute of Technology & Research, Lovely Professional University, Phagwara,
Badnera, Amravati, India Punjab, India.
rokadeashay@[Link] manwinder.25231@[Link]

Anudeep Goraya Yogesh Thakare

School of Electronics and Electrical Engineering Computer Science & Engineering (AIML)
Lovely Professional University, Phagwara, Shri. Ramdeobaba College of Engineering and
Punjab, India. Management, Nagpur, India.
[Link]@[Link] thakareyn@[Link]

Abstract—The Internet of things (IoT) allows for previously Understanding and managing the inherent unpredictability
impossible interaction between previously unlike objects. The within a field is where precision agriculture shines, allowing
implementation of IoT is essential to the success of the agricultural for dramatically increased productivity. To improve
industry. Using these techniques, farmers can increase their crop agricultural production and facilitate strategic decisions,
yields while using fewer resources. However, most farmers still lack researchers have been studying precision agriculture (PA),
knowledge of modern techniques. Using IoT capabilities, precision,
and intelligent agriculture manages agricultural production to
which involves the application of cutting-edge sensors and
improve crop quality by delivering the necessary nutrients. It reduces analysis tools. Precision agriculture is being used to combat
environmental damage caused by the use of excessive pesticides. In late-blight disease through machine learning. The algorithm
this research, the IoT and ML is used as a foundation to investigate uses information about the plant's environment and climate
the agricultural sustainable development platform to enhance the to recommend preventive measures to the farmer. In order to
platform's management effect. This research described a precise and manage continuous field monitoring more effectively, IoT is
intelligent agricultural system that uses supervised classification and also used to collect sensor data from the field, making the
regression-based machine learning to analyse predicted data on data and the guidance of an ML algorithm accessible on a
sensing parameters. The fog, edge, and sensor layers are the three GUI platform. The main takeaways are that an IoT platform
crucial parts of the suggested methodology. For controlling the
actuators in the system, the application of machine learning on sensor
may help farmers boost output and that agricultural needs
data gathered from prototype embedded models is being examined. can be predicted using historical data [2]. Due to the
Then, at the fog layer, an analytics and decision-making system was widespread use of this technology, increased crop output,
developed using two supervised machine learning techniques, decreased waste, enhanced pest control, and simplified
including classification and regression algorithms that made use of animal care are just some benefits farmers may expect to
decision trees (DT) and multilayer perceptron neural networks enjoy. Multisource data is used in precision agriculture to
(MLPNN) for efficient processing. When the experimental findings maximise yields while decreasing the overall cost of inputs,
are examined and analysed, it is discovered that DT has a including fertiliser, water, and pesticides. Domestic
significantly higher classification accuracy than MLPNN and other subfield-scale technology and methods are what is meant by
cutting-edge techniques.
the term "precision agriculture” [3].
Index Terms— Internet of Things, Intelligent Farming, Machine
Learning, Precise Farming, Supervised Modelling.
II. RELATED WORK
This paper provides an overview of predictive analysis,
I. INTRODUCTION IoT designs accompanying cloud presidency, and security
To maximise profits, sustainability, and environmental flaws for multibreed in the farming industry [4] considering
protection, modern farms are adopting a novel idea known as farmers' historical experiences. It also emphasises the
"intelligent and precision agriculture," which is characterised difficulties and difficulties that will be encountered when
as an information technology-based farm management system conventional farming practices are combined with
that uses field variability detection, analysis, and control. In contemporary technologies. It is suggested [6] to integrate
the same way, field crops need pesticides, manures, and a an intelligent production system with an effective prediction
water system for significantly better development; precision approach called WPART produced by machine intelligence
agriculture (PA) could be applied for the same reason in techniques to think about crop productivity and dryness in
cultivation. Better judgments can be made in many areas of order to provide an acceptable decision support in intelligent
crop production using modern information technologies, farming systems. As predicted, IoT and in-depth knowledge
which is what precision agriculture is all about [1].[Link] are combined in a system that will randomly monitor paddy
979-8-3503-6052-3/24/$31.00 ©2024 IEEE

Authorized licensed use limited to: Lovely Professional University - Phagwara. Downloaded on July 31,2024 at [Link] UTC from IEEE Xplore. Restrictions apply.
fields [7]. For rice leaf disease detection and nitrogen rank • Create a three-tier architecture for an intelligent farming
belief, the preprepared network vgg16 is being analysed. system to make it easier to implement a low-cost farming
Data from several crops were gathered by a sensor network system with intelligent solutions.
[8]. Numerous machine learning approaches have been • Run a number of tests to evaluate the effectiveness of the
developed to predict plant infestation with CLW [9]. An suggested analytics and decision-making model in light of
exhaustive study of smart agricultural approaches and the outcomes of supervised machine learning.
frameworks is proposed [10]. Applications of intelligent
sensor systems enabled by machine learning are described III. PROPOSED MODEL
[11]. For a number of material limits with strong The proposed paradigm for an exact and intelligent
nonlinearity and turbulence, we provide [12] a bidirectional greenhouse farming system is shown in Fig. 1. The fog
self-concentrating encoder-translator framework (BEDA). layer, the edge layer, and the sensor/device layer are the
An indicator model for apple rot was given in the orchards three main architectural layers. All field environment-related
of the Kashmir basin using an IoT system using secondary sensors and actuators are housed in the sensor layer. For
data sources and AI [13]. data collection and transmission to the fog layer, the edge
Intelligent agriculture settings are offered [14], and their layer typically consists of a control unit to which sensors
varied architectures, applications, and research calendars. In and actuators are linked. One of the main responsibilities is
[15], the authors provide an ecosystem overview of the to produce the control signals that the edge layer will use to
ecosystem of IoT and show how DA and IoT work together regulate the actuators based on the analytics and decision-
to make intelligent agriculture a reality. An approach called making model that the fog layer has created using the data it
Intelligent Agriculture [16] has been developed to allow has gathered from the edge layer.
farmers to considerably boost output by monitoring A) Sensor Layer
agricultural fields. An IoT-based creative agricultural A low-power microcontroller that is installed in the field and
listening system will be accompanied by a number of is specifically designed for the Internet of Things connects the
discoveries, quantification, maturity testing, and unhealthy nodes and edges. The experimental plants chosen for this
produce detection algorithms [17]. Benefits include a greenhouse experiment were gerbera daisies and broccoli.
decrease in labour expenses and the creation and use of a This type of crop can benefit greatly from the use of a
cutting-edge wireless mobile robot that makes use of the IoT greenhouse management system, which controls relative
to perform numerous outdoor operations with better humidity (RH), temperature (T), and light (L), as well as
accuracy and efficiency. A framework emerged [19-20] that protects the crops from precipitation (rain, storms, and
incorporates IoT-accompanying crop outcomes, using cloud- scorching sun) and eradicates pests and diseases.
based estimation to keep an eye on crops while using B) Edge Layer
various units of measurement and instructions. The An edge node in our experiment is a Node MCU ESP 32,
following are the most significant contributions. which can collect data from sensors and process it locally
• Use supervised machine learning to create a powerful
analytics and decision-making model for precision and
intelligent farming.

Fig.1 Precise and intelligent smart farming framework

Authorized licensed use limited to: Lovely Professional University - Phagwara. Downloaded on July 31,2024 at [Link] UTC from IEEE Xplore. Restrictions apply.
before sending it to the base station of the edge layer. Sensors unpredictability in the subsets. The objective is to identify the
need to be calibrated and verified against an expected value in feature that optimises either more information or less
order to obtain analogue or digital data that comply with impurity, which will arise from the split.
standards. Data collection for a wide range of favorable and
unfavorable environmental circumstances is important to
ensure crop survival through precision crop management. The
hardware interface of the proposed experimental model is
shown in Fig. 2. Sensor data are viewed in real time via a
controller on a personal computer and then uploaded to the
cloud.

Fig.2 Proposed experimental setup

C) Fog Layer Fig. 3 Model working flow

The main purpose of the analytics and decision-making model
∑
is to control activity at the edge layer and transmit pertinent
(1)
data to the cloud layer, where farmers can access it. The
authoritative sequence will create a machine intelligence
1 ∑ (2)
treasure with different processing stages. The suggested
strategy was developed in the fog layer and relies heavily on
Where,
machine learning techniques, especially when it comes to
classification modelling and the analytics system. The Internet pi is the proportion of elements in the set that belong to the ith
of Things (IoT) devices, particularly sensors (temperature, category.
humidity, camera vision, light intensity, etc.), collect data in Multilayer perceptual networks are multilayered neural
real-time or in tiny batches. (When to start a water pump, networks. We can construct a neural network by connecting
etc.). Decision-making based on predictions and data neurones together, where some neurones' outputs serve as
visualisation through dashboards and reports. Algorithms like inputs to other neurons. The sigmoid activation function is u,
decision trees (DT) and multilayer perceptron neural networks
(MLPNN) are used to create predictive models. 1, !0
The Decision Tree method selects the proper characteristic 0, ℎ $%
to split the data during training based on a metric such as
entropy shown in equation 1 or Gini impurity shown in send at each node in multilayer perception. The sigmoid
equation 2, which evaluates the degree of impurity or activation function uses the sigmoid formula to transform real
input values into a value between zero and one. A weighted

Authorized licensed use limited to: Lovely Professional University - Phagwara. Downloaded on July 31,2024 at [Link] UTC from IEEE Xplore. Restrictions apply.
sum of the input features is calculated by passing the features
on to an input function u, as shown in Equation 3.
∑ $ (3)

By applying the outcome of this calculation to an activation

function f, the perceptron produces its output. The activation
function of the classic perceptron is a step function, as shown
below in Equation 4.

1, !0
(4)
0, ℎ $%

Where; θ is a threshold parameter.

IV. EXPERIMENTAL RESULTS AND DISCUSSION

A) Experimental Setup
On a laptop the analytics and decision-making model Fig. 4(a) Temperature from experiment setup for season 1
classification was carried out with no other background
processes active. Tools from MATLAB's statistics and
machine learning toolbox were used to programme the
intelligent model. System performance is assessed using a
confusion matrix as a performance metric. Additionally, the
parameters of the confusion matrix are used to determine the
accuracy, recall, precision, f score, negative predictive value
(NPV), false negative rate (FNR), and false positive rate
(FPR) for the classification model and RMSE for the
regression model. True Positive, False Positive, True
Negative, and False Negative are the four types of results.

B) Data Acquisition
The proposed experimental strategy is put into practice on a
prototype and tested under varied conditions on two crops,
gerbera and broccoli. The obtained data set includes the
following feature attributes: CO2 gas level (ppm), soil
wetness (percent), light intensity (lux), humidity (percent) and
temperature (Celsius) with the following output attributes: Fig.4(b) Humidity monitoring from experiment setup for season 1
pump (on/off) ventilation fan (on/off), and amount of light
(low/medium/high). The data sets’ values were collected over
a period of days and hours under diverse settings. Fig. 4
(a,b,c,d) display the average of each day's readings for each
sensor's attribute during summer season. By publishing data
from the nodes to the Adafruit broker, all of this sensor data
may be viewed online on the Adafruit IO Cloud dashboard.
The user can then subscribe to these data to obtain it
immediately.
Depending on soil moisture levels, the greenhouse doors
and windows can easily be opened and off. Instead of during
During the day, the plant photosynthesis process requires high
levels of CO2 concentration and water in the evenings. With
these two energy sources, the photosynthesis process keeps
the plant cool and promotes quick growth. After conducting a
CO2 concentration level experiment in a greenhouse, the
greenhouse maintains a maximum CO2 level at night because Fig.4(c) CO2 monitoring checking from experiment setup for season 1
it absorbs CO2 from day to night.

Authorized licensed use limited to: Lovely Professional University - Phagwara. Downloaded on July 31,2024 at [Link] UTC from IEEE Xplore. Restrictions apply.
 TABLE II
PERFORMANCE EVALUATION OF THE CLASSIFICATION MODEL

NPV (%) FNR (%) FPR (%)

Attributes
DT MLPNN DT MLPNN DT MLPNN
Pump 8.55 12.78 0 21.76 5.16 15.55
Light 10.45 24.46 7.20 61.55 6.15 27.2
Fan 7.86 18.90 18.59 17.46 11.22 14.5

For the DT and MLP-NN methods, the simulation results for

regression modelling are assessed and examined, respectively.
These findings show the results of the regression report for the
output attributes pump, fan, and light. Table 4 shows the
effectiveness of the regression model using the RMSE metric.
Fig.4(d) Soil moisture acquisition from experiment setup for season 1 This table reveals that the DT regressor technique outperforms
the MLPNN regressor for all three output qualities.

TABLE III
C) Performance Evaluation of Classification- and PERFORMANCE EVALUATION OF THE CLASSIFICATION MODEL
Regression-Based Modelling
Attributes Pump Light Fan
The confusion matrix for regression and classification model
has been tested to acquire the intended result and calculate DT 89.86 95.68 93
various metrics such as accuracy, sensitivity, latency, etc. Accuracy (%) MLPNN 77.86 72 84.53
after applying the suggested methods to the standard data set.
DT 83.81 100 92.44
MATLAB software with statistics and a machine learning Recall (%) MLPNN 79.35 90.27 84.41
toolbox was used to train and evaluate the model using the
two supervised machine learning models, DT and MLPNN. DT 100 92.80 81.41
The grid search CV technique was used to find the best Precision (%) MLPNN 78.24 38.45 82.54
hyperparameters that were needed for both algorithms to train DT 94.84 93.85 88.78
the model. The simulation results for classification modelling F-score (%)
are evaluated for the DT and MLPNN algorithms, MLPNN 84.45 72.80 85.50
respectively. In these results, the latency evaluation for the
training and testing phase of both classification- and
regression-based models is shown in Table 3. It is observed TABLE IV
PERFORMANCE EVALUATION OF THE REGRESSION MODEL
that the MLP-NN algorithm takes more time to train and test
the model as compared to DT. RMSE
Attributes
TABLE I DT MLPNN
LATENCY EVALUATION OF THE CLASSIFICATION AND REGRESSION MODEL

Training Time (sec) Testing Time (sec) Pump 0.0701 0.0076

Classifiers
DT MLPNN DT MLPNN Fan 0.0053 0.0022
Classification
15.5 21.7 5.3 5.9 Light 0.0262 0.0128
Modelling
Regression
16.1 22.4 4.7 5.2 D) State-of-the-Art Comparison
Modelling
Table 5 concludes by comparing the methodologies with the
Tables 2 and 3 present the findings of the classification report latest ones. Considering the parameters of the classification
of the classification-based modelling for the three output model of accuracy, recall and f score and the parameters of
qualities of the pump, fan, and light. This table shows that, for the regression model parameters of RMSE, it is concluded
all three output qualities, the DT method outperformed the from the results that the proposed classification and regression
MLPNN in all performance criteria. model, which is used for intelligent and precise smart farming
for the greenhouse model, produces better results.

Authorized licensed use limited to: Lovely Professional University - Phagwara. Downloaded on July 31,2024 at [Link] UTC from IEEE Xplore. Restrictions apply.
 TABLE V [5] Lakshmanna, K. et la., “Deep Learning-Based Privacy-Preserving Data
COMPARATIVE ANALYSIS Transmission Scheme for Clustered IIoT Environment”, Computational
Reference Accuracy Recall F-Score Latency Intelligence and Neuroscience. 2022 pp. 1-11, June 2022.
RMSE [6] Rezk, N. et al., “An efficient IoT based smart farming system using
Model (%) (%) (%) (sec)
machine learning algorithms”, Multimedia Tools and Applications. 80,
[4] - 93.59 96.30 - - 773-797, Sept. 2020.
[5] - - - 0.2431 - [7] Tageldin, A. et al., “Applying Ma- chine Learning Technology in the
[6] 87.35 87.33 87.17 - 224.6 Prediction of Crop Infestation with Cotton Leafworm in Greenhouse” ,
Cold Spring Harbor Laboratory, Oct. 2020.
[10] 90 - - - - [8] Aliar, A. et al., “A comprehensive analysis on IoT based smart farming
[18] - - - 0.02726 - solutions using machine learning algorithms”, Bulletin of Electrical
Our Engineering and Informatics. 11, 1550- 1557, June 2022.
95.84 94.75 96.49 0.0338 6.49 [9] Sethy, P. et al., “Smart paddy field monitoring system using deep
Work learning and IoT”, Concurrent Engineering. 29, 16-24, Jan. 2021.
[10] Araby, A. et al, “Smart IoT Monitoring System for Agriculture with
Predictive Analysis”, 2019 8th International Conference on Modern
V. CONCLUSIONS Circuits and Systems Technologies (MOCAST), May 2019.
[11] Nam, Ha et al., “Machine Learning- Enabled Smart Sensor Systems.
In this research, an IoT-enabled, machine learning-based, Advanced Intelligent Systems”, In Advanced Intelligent Systems, Vol.
precise, and intelligent agricultural system was presented. The 2, Issue 9, July 2020.
suggested method can be used in an IoT-based decision- [12] Jin, X. et al., “Deep- Learning Temporal Predictor via Bidirectional
making paradigm in a smart agricultural setting. Two Self-Attentive Encoder–Decoder Framework for IOT-Based
Environmental Sensing in Intelligent Greenhouse”, Agriculture. 11, 802,
greenhouse plants have their sensing parameters entered into Aug. 2021.
an IoT based embedded system device, and the relevant [13] Akhter, R. & Sofi, S. “Precision agriculture using IoT data analytics and
actuators get output from the system. Classification and machine learning”, Journal of King Saud University - Computer and
regression models are the two types of analytics used to create Information Sciences. 34, 5602-5618, Sept. 2022.
[14] Quy, V. et al., “A. IoT-Enabled Smart Agriculture: Architecture,
programmes for smart and accurate farming. Both DT and Applications, and Challenges”, Applied Sciences. 12, 3396, March
MLPNN can be utilised for these modelling purposes. Finally, 2022.
classification- and regression-based supervised machine [15] Elijah, O. et al., “An Overview of Internet of Things (IoT) and Data
learning algorithms are used to evaluate how well the smart Analytics in Agriculture: Benefits and Challenges”, IEEE Internet of
Things Journal. 5, 3758-3773, Oct. 2018.
farming system performs in intelligent and precise farming. [16] Thakare, Y. et al., “Intelligent Life Saver System for People Living in
The results demonstrated that DT outperformed MLPNN and Earthquake Zone”, International Journal of Next-Generation
other cutting-edge classification algorithms. Using the Computing, Feb. 2023.
suggested approach, smart agricultural systems with [17] Noor-Alam Siddiquee, K. et al., “Development of Algorithms for an
IoT-Based Smart Agriculture Monitoring System”, Wireless
intelligent prediction-based decision support can be created Communications and Mobile Computing. 2022 pp. 1-16, April 2022.
with comparable degrees of success. The suggested technique [18] Ferehan, N. et al., “Smart Farming System Based on Intelligent Internet
also provided the most accurate predictions for actuators and of Things and Predictive Analytics”, Journal of Food Quality. 2022 pp.
the most precise control. 1-8, May 2022.
[19] Rokade, A. & Singh, M. “Analysis of Precise Green House
Additionally, by making this suggested model durable and Management System using Machine Learning based Internet of Things
waterproof, it can be applied in real-world circumstances. (IoT) for Smart Farming”, 2021 2nd International Conference on Smart
Additionally, the analytics model will change depending on Electronics and Communication (ICOSEC), Oct. 2021.
the crop or plant considered in some cases. Deep learning [20] Rokade, A. et al., “Intelligent Data Analytics Framework for Precision
Farming Using IoT and Regressor Machine Learning Algorithms”,
models with large data set samples can be used to enhance the Applied Sciences. 12, 9992, Oct. 2022.
model's performance. The analytics methodology will also [21] Rokade, A. et al., “IOT-Based Medical Informatics Farming System
change depending on the crop or plant under investigation. with Predictive Data Analytics Using Supervised Machine Learning
Using deep learning models trained on massive dataset Algorithms”, Computational and Mathematical Methods in Medicine.
2022 pp. 1-15, Aug. 2022.
samples may enhance model performance. [22] Rokade, A. & Singh, M. “Smart Farming System Based on IoT for
Precision Controlled Greenhouse Management”, Lecture Notes in
REFERENCES Electrical Engineering. pp. 435- 443, 2023.

[1] J. Pei et al., "TKAGFL: A Federated Communication Framework Under

Data Heterogeneity," in IEEE Transactions on Network Science and
Engineering, vol. 10, no. 5, pp. 2651-2661, 1 Sept.-Oct. 2023.
[2] Lakshmanna, K et al., “Improved Metaheuristic-Driven Energy-Aware
Cluster-Based Routing Scheme for IoT-Assisted Wireless Sensor
Networks”, Sustainability 14(13), 7712, June 2022.
[3] V, Suma. “Internet of Things (IoT) based Smart Agriculture in India:
An Overview”, Journal of ISMAC. 3, 1-15 Feb.2021.
[4] Gundluru, N. et al., “Enhancement of Detection of Diabetic Retinopathy
Using Harris Hawks Optimization with Deep Learning Model”,
Computational Intelligence and Neuroscience. 2022 pp. 1-13, May
2022.

Authorized licensed use limited to: Lovely Professional University - Phagwara. Downloaded on July 31,2024 at [Link] UTC from IEEE Xplore. Restrictions apply.