See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/372805391

Machine Learning Algorithms for Prediction of Mobile Phone Prices

Chapter · August 2023

DOI: 10.1007/978-981-99-3010-4_7

CITATIONS READS

2 507

4 authors, including:

Jinsi Jose
Christ University, Bangalore
10 PUBLICATIONS 54 CITATIONS

SEE PROFILE

All content following this page was uploaded by Jinsi Jose on 20 February 2024.

The user has requested enhancement of the downloaded file.

 Machine Learning Algorithms for Prediction of Mobile
Phone Prices

Jinsi Jose1* [0000-0002-4691-6474], Vinesh Raj1, Sweana Vakkayil Seaban1, and Deepa V
Jose2[0000-0001-2651-6374]
1Dept. of Computer Science, Rajagiri College of Social Sciences, Kalamassery, India
2Dept. of Computer Science, Christ University, Bangalore, India

jinsijose@rajagiri.edu

Abstract. The drastic growth of technology helps us to reduce the man work in
our day-to-day life. Especially mobile technology has a vital role in all areas of
our lives today. This work focused on a data-driven method to estimate the
price of a new smartphone by utilizing historical data on smartphone pricing,
and key feature sets to build a model. Our goal was to forecast the cost of the
phone by using a dataset with 21 characteristics related to price prediction. Lo-
gistic Regression (LR), Decision Tree(DT), Support Vector Machine (SVM),
Naive Bayes algorithm (NB), K-Nearest Neighbor (KNN) algorithm, XGBoost,
and AdaBoost are only a few of the popular machine learning techniques used
for the prediction. The Support Vector Machine achieved the highest accuracy
(97%) compared to the other four classifiers we tested. K-Nearest Neighbor’s
94% accuracy was close to that of the support vector machine.

Keywords: Mobile phone, phone price prediction, machine learning, support

vector machine, K-Nearest Neighbor, Price range.

1 Introduction

An electronic gadget that is portable and linked to a cellular network is referred to as

a "mobile phone." In 1973, Motorola created the first portable cell phone. According
to Cisco Annual Internet Report (2018–2023), over 70 percent of the global popula-
tion will have mobile connectivity by 2023. The total number of global mobile sub-
scribers will grow from 5.1 billion (66 percent of the population) in 2018 to 5.7 bil-
lion (71 percent of the population) by 2023 [1]. Mobile phones are designed to allow
people to communicate via phone and email. In addition to making phone calls and
sending emails, people can now access the internet, send text messages, and play
games. Mobile phones are equipped with a variety of technologies and features. The
most recent mobile phone technologies are WIFI, high-quality cameras, more cores,
and large memory capacities. The performance of the mobile phone varies depending
on the features used. As performance improves, so does the price range, which shifts
from low to high.
The primary and foremost aim of the work is to predict the mobile phone price
based on attributes considered as the specification for mobile phone usage. It was the
2

right approach to predict the future prices of smartphones accurately. Customers and
business owners benefit from this information when purchasing a phone. Predicting
the pricing will allow consumers to make more informed judgments when selecting a
new phone. A few features considered the smartphone's features are display, pro-
cessing power, memory, camera, thickness, battery life, and connectivity. A commod-
ity's intrinsic value is often misunderstood. Inadequate tools for cost-benefit analysis
lead to poor decision-making.
In today's world, being without a smartphone is practically impossible. In an earli-
er technological revolution, mobile phones were used only for communicating with
others during our mobility. The first mobile phone was Motorola DynaTAC 8000X,
launched in 1973 by Dr Martin Cooper [2][3]. The first product needed ten hours to
charge and weigh a kilogram. In the present scenario, from this basic structure of the
functioning mobile phones have a vital role in everybody’s life worldwide. The out-
come of the momentous development of technology leads to more involvement of
mobile phones in our day-to-day life. Because it is a portable device capable of bring-
ing everything to our fingertips in a fraction of a second. The mobile phone has vari-
ous applications, such as education, business, banking, entertainment, etc. Even
though the mobile phone has various applications, the features are very important in
pricing [4].
Machine Learning (ML) is a subset of artificial intelligence capable of performing
as human intelligence. Artificial intelligence systems are used to solve complex tasks
efficiently. Machine learning methods adopted various methods based on learning
capabilities to learn and analyse the problem. The different machine learning tech-
niques are supervised, unsupervised, semi-supervised and reinforcement learning to
solve real-world problems [5][6].
This article focus on predicting the price range of mobile phones using twenty dif-
ferent attributes of a dataset that are features of various mobile phones used world-
wide. According to the mobile phone properties, a phone is classified into one of four
price ranges ranging from zero to three. Where zero denotes a low-budget mobile, one
denotes an upper middle-budget mobile, two denotes a middle-budget mobile, and
three denotes a high-budget mobile.
The structure of the paper has organized as follows: features and relevance of ma-
chine learning techniques in mobile phone price prediction and various methods im-
plemented by various researchers are given in Section 2. Section 3 gives an idea about
the implemented methodologies and describes the different model building, followed
by results and discussion in Section 4. Finally, Section 5 concludes the paper.

2 Literature Review

Analyzing the previous data and predicting the future of the upcoming product is
unavoidable in every machine-learning research. The researchers worked on different
machine-learning algorithms for mobile price prediction based on feature selection
methods [7]. This work identified a better feature selection algorithm and good classi-
fier to get higher accuracy. From the comparison, the result can be concluded that the
 3

Decision Tree (DT) classifier achieved 87% of maximum accuracy. Another study
carried out by I. Nasser et. al., predicted the mobile phone price range by using Artifi-
cial Neural Networks (ANN) [8]. After the training and validation, the model yielded
an accuracy of 96.31%.
Another study explained the prediction price using three classifiers: random lo-
gistic regression and SVM [9]. In terms of accuracy, researchers concluded was the
best classifier with 81% accuracy. A study was carried out by P. Arora et. al. on a
prediction model using the WEKA tool [10]. The researchers implemented ZeroR
algorithm, Naïve Bayes (NB) algorithm, and J48 decision tree algorithm. The results
have shown J48 decision tree algorithm achieved better accuracy.
Another work done is developing machine learning models for the prediction of
new mobile phone prices by using Support Vector Machine (SVM), Random For-
est(RF) classifier, and Logistic regression (LG) [11]. By the analysis of the result
understood that SVM achieved a high accuracy with 97% rather than the other two
classifiers. The researchers K. Karur and K. Balaje presented K-Nearest Neighbor
(KNN) for predicting mobile phone prices [12]. In this work, researchers were fo-
cused on feature selection and based on ram size decided the phone's price range. The
researchers implemented six machine-learning algorithms for price prediction. The
researchers used the ANOVA f-test for the feature selection, and the Linear Support
Vector Machine (SVC) yielded high accuracy in price prediction [13].
Another study used supervised machine learning algorithms for price prediction
[14]. The researchers considered the confusion matrix and accuracy as the evaluation
metrics. Compared to other supervised classifiers Linear Discriminant Analysis
(LDA) achieved high accuracy with 95%. In another study, researchers focused on a
hybrid model for mobile price prediction. The authors implemented the decision tree
and random forest method and achieved 83% and 84% accuracy respectively [15].

3 Materials and Methods

3.1 Dataset
The dataset was about mobile prices across different areas of the world. It contains 21
attributes. The attributes are the details of mobile phones like Battery power, internal
memory, ram capacity, price range, and all. The source of the dataset is Kaggle. Every
attribute in the dataset has been used to classify the data. Ram indicates the ram ca-
pacity of the mobile phone, and the price range varies in four ranges from zero to
three. we are classifying the price range by considering other attributes, and we will
be predicting the price range after the training of the dataset. Fig. 1 shows the correla-
tion between features in the dataset.

3.2 Pre-processing
Data pre-processing is the method by which the raw data is into a robust, understand-
able format. Data in the raw format is frequently inconsistently formatted, contains
human errors, and may be incomplete. Such issues are resolved by data pre-
4

processing, which makes datasets completer and more efficient for data analysis. It is
an important step that can influence the success of projects involving data mining and
machine learning. It speeds up knowledge discovery from datasets and may eventual-
ly affect the performance of machine learning models. After pre-processing, the above
dataset was divided into train and test data. Train data and test data are in the ratio of
80:20.

Fig. 1. Correlation of attributes.

3.3 Model building

Seven machine learning algorithms obtained the price range prediction from the given
attribute; AdaBoost, Decision Tree, K-Nearest Neighbor, Logistic regression, Naïve
Bayes, Support Vector Machine, and XGBoost in data mining. The models are evalu-
ated using accuracy, precision, recall and f1-score.
AdaBoost. It is also known as the Adaptive Boosting algorithm. It follows an ensem-
ble learning methodology, building a robust classifier from weak ones. In this model,
iteratively build models until the model becomes free of errors. To improve classifier
accuracy, the AdaBoost Classifier combines many classifiers. The AdaBoost classifier
combines several weak classifiers to create a strong, accurate classifier. It has a 57%
accuracy rate. Table 1 gives the evaluation metrics of the AdaBoost algorithm.

Table 1. Classification report of AdaBoost.

Evaluation Metrics Accuracy Precision Recall F1-score

Percentage 57 69 59 55
 5

Decision tree classifier. The decision tree follows a supervised machine-learning

approach. This algorithm solves problems using tree representation. Each leaf node is
a class label in this representation, and the internal node represents attributes. Deci-
sion nodes are the points at which data is split. This approach can be used to solve
classification and regression problems. This algorithm aims to create a training model
that can predict the class range of a target variable with the aid of simple decision
rules learned from training data. The accuracy after predicting the price range is 82%.
Table 2 shows the evaluation metrics of the decision tree classifier.

Table 2. Decision Tree classifier

Evaluation Metrics Accuracy Precision Recall F1-score

Percentage 82 88 90 89

K-Nearest Neighbors Algorithm. A convenient supervised machine-learning classi-

fication approach is K-Nearest Neighbours. It categorises a data point based on its
nearness to its neighbors. Parameter tuning is the technique of choosing an appropri-
ate value for K. The principle of the KNN algorithm is that a new data point falls into
the Class of points to which it is close. There is no specific method for determining
the optimal K value. It is determined by the type of problem at hand and the business
scenario. Five is the most preferred value for K. Choosing a K value of one or two can
be noisy and result in outliers in the model, resulting in overfitting. The algorithm
performs well on the training set compared to its true performance on unseen test
data. To predict the values after fitting the trained data and running, the algorithm
yielded 94% accuracy. Table 3 shows the KNN classification report.

Table 3. Classification report of KNN.

Evaluation Metrics Accuracy Precision Recall F1-score

Percentage 94 97 94 95

Logistic Regression. A fundamental classification technique is logistic regression. It

is associated with polynomials and linear regression and is a member of the linear
classifier family. The logistic regression findings are simple to interpret, and the pro-
cess is rapid and simple to comprehend. It may be used to solve multiclass issues even
though it is a binary classification approach. The accuracy of the prediction was 63%.
Table 4 given the evaluation metrics of the logistic regression.

Table 4. Evaluation metrics of logistic regression.

Evaluation Metrics Accuracy Precision Recall F1-score

Percentage 63 69 73 71

Naïve Bayes. Instead of supplying a test point's label, the Naive Bayes classifier algo-
rithm returns the likelihood that it belongs to a certain class. This model is one of the
6

simplest Bayesian network models. However, it can attain better levels of accuracy
when paired with kernel density estimation. This method is exclusively appropriate
for Classification jobs, unlike many other ML algorithms, which can often handle
both Regression and Classification tasks.
The naïve Bayes method is so called because it is practically hard to establish evi-
dence for its assumptions in empirical data. To get the sum of the component proba-
bilities, conditional probability is used The accuracy calculated by Naïve Bayes is
80%. Table 5 shows the evaluation metrics of Naïve Bayes algorithm.

Table 5. Evaluation metrics of Naïve Bayes.

Evaluation Metrics Accuracy Precision Recall F1-score

Percentage 80 88 81 85

Support Vector Machine. Support Vector Machines, is an old, well-known, and

sophisticated algorithm. The SVM classifier is widely regarded as one of the best
linear and non-linear binary classifiers available. SVM regressors are also becoming
popular as an alternative to traditional regression algorithms like Linear Regression.
In N-dimensional space, the SVM method seeks a hyperplane that can distinguish
between data points (N — the number of features). Various hyperplanes are used to
separate two classes of data points. The largest margin or distance between data
points from both classes is what we are looking for in a plane. Increasing the margin
distance, makes it possible to classify next data points with more assurance. Table 6
given the classification of SVM.

Table 6. Classification of SVM

Evaluation Metrics Accuracy Precision Recall F1-score

Percentage 97 98 99 99

XGBoost. XGBoosting stands for extreme Gradient Boosting, in which each predic-
tor corrects its predecessor's error. Here decision trees are created in sequential form.
It falls under boosting ensemble learning. Artificial neural networks often outperform
all other algorithms or frameworks for unstructured data prediction issues. 89.5%
accuracy is acquired via XGBoost. Table 7 shows the classification report of
XGBoost algorithm.

Table 7. Classification of XGBoost.

Evaluation Metrics Accuracy Precision Recall F1-score

Percentage 90 94 90 92
 7

4 Experimental Results and Discussion

We used several algorithms to make our price range predictions, including the Sup-
port vector machine, the Decision tree, naïve Bayes, the K-nearest neighbor, and lo-
gistic regression algorithms. The most accurate prediction was made by the support
vector machine method (97%) and the K-nearest algorithm (94%). Predictions of
future prices were also made using XGBoost (89.5%), decision tree (82%), a naive
Bayes model (80%), logistic regression (63%) and AdaBoost (57%). The performance
of the given algorithms is shown in Fig. 2.
Comparison of Accuracy
120
94 97
100 89.5
80 82
80
63
57
60

40

20

Fig. 2. Accuracy comparison of different models.

Therefore, the support vector machine algorithm can be considered the most effec-
tive for this task. This algorithm considers a wide variety of mobile phone characteris-
tics, including the storage capacity, the number of processor cores, the battery life,
and so on, to make an accurate price prediction. After achieving a 98% accuracy rate
on the training data and a 97% accuracy rate on the test data, we can confidently de-
clare that our method is effective.

5 Conclusion

The strategies utilized in this article to estimate the price range include the Support
vector machine, K-nearest neighbor algorithm, decision tree, naive Bayes algorithm,
and logistic regression. The support vector machine algorithm had the highest predic-
tion accuracy of 97%, while the K-nearest algorithm had 94%, which was closest to
the SVM algorithm. Other price prediction algorithms, such as decision tree, naive
Bayes, and logistic regression, achieved 82%, 80%, and 63% accuracy, respectively.
 8

References
1. Cisco Annual Internet Report - Cisco Annual Internet Report (2018–2023) White Paper –
Cisco, https://www.cisco.com/c/en/us/solutions/collateral/executive-perspectives/annual-
internet-report/white-paper-c11-741490.html, last accessed 2022/11/26.
2. R. Hossain, M. R. Hasan, and M. Sharmin.: A Short Review on the History of Mobile
Phones. Journal of Android, IOS Development and Testing 7(2), 33–39 (2022).
3. Evolution of Smartphone, https://www.researchgate.net/ publication/355041882_Evolution
_of_Smartphone, last accessed 2022/11/30.
4. E. Poppe, M. Jaeger-Erben, and M. Proske.: The smartphone evolution-an analysis of the
design evolution and environmental impact of smartphones. In Electronics Goes Green
2020, Berlin, pp. 1–9. Germany (2020).
5. I. H. Sarker.: Machine Learning: Algorithms, Real-World Applications and Research Di-
rections. SN Computer Science 2(160), 4-21 (2021). doi: 10.1007/s42979-021-00592-x.
6. K. G. Liakos, P. Busato, D. Moshou, S. Pearson, and D. Bochtis.: Machine learning in ag-
riculture: A review. Sensors (Switzerland) 18(8) 1–29, (2018). doi: 10.3390/s18082674.
7. M. Asim and Z. Khan.: Mobile Price Class prediction using Machine Learning Tech-
niques. International Journal of Computer Applications 179(29), 6–11(2018). doi:
10.5120/ijca2018916555.
8. I. M. Nasser and M. Al-Shawwa.: Developing Artificial Neural Network for Predicting
Mobile Phone Price Range. International Journal of Academic Information Systems Re-
search 3(2) 1–6, (2019).
9. Subhiksha, S., Thota, S., Sangeetha, J. (2020). Prediction of Phone Prices Using Machine
Learning Techniques. In: Raju, K., Senkerik, R., Lanka, S., Rajagopal, V. (eds) Data Engi-
neering and Communication Technology. Advances in Intelligent Systems and Computing,
vol 1079. Springer, Singapore. https://doi.org/10.1007/978-981-15-1097-7_65.
10. P. Arora, S. Srivastava, and B. Garg.: Mobile Price Prediction Using Weka. International
Journal of Scientific Development and Research 5(4) 330–333, (2020).
11. K. S. Kalaivani, N. Priyadharshini, S. Nivedhashri, and R. Nandhini.: Predicting the price
range of mobile phones using machine learning techniques. In: AIP Conference Proceed-
ings, (2021). doi: 10.1063/5.0068605.
12. K. Karur and K. Balaje.: Prediction of Mobile Model Price using Machine Learning Tech-
niques International Journal of Engineering and Advanced Technology. 11(1) 273–275,
(2021). doi: 10.35940/ijeat.a3219.1011121.
13. M. Cetin and Y. Koc.: Mobile Phone Price Class Prediction Using Different Classification
Algorithms with Feature Selection and Parameter Optimization. In: ISMSIT 2021 - 5th In-
ternational Symposium on Multidisciplinary Studies and Innovative Technologies, Pro-
ceedings, pp. 483–487. IEEE, Turkey (2021). doi: 10.1109/ISMSIT52890.2021.9604550.
14. A. Varun Kiran.: Prediction of Mobile Phone Price Class using Supervised Machine
Learning Techniques. International Journal of Innovative Science and Research Technol-
ogy 7(1) 248-251, (2022).
15. A. H. Sakib, A. K. Shakir, S. Sutradhar, M. A. Saleh, W. Akram, and K. B. M. B. Biplop.:
A hybrid model for predicting Mobile Price Range using machine learning techniques. In:
ACM International Conference Proceeding Series, pp. 86–91. Association for Computing
Machinery, Thailand (2022). doi: 10.1145/3512850.3512860.

View publication stats