Prediction of Mobile Phone Price Class Using Supervised Machine Learning Techniques
Prediction of Mobile Phone Price Class Using Supervised Machine Learning Techniques
Volume 7, Issue 1, January – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Prediction of Mobile Phone Price Class using
Supervised Machine Learning Techniques
A Varun Kiran and Dr. Jebakumar R.
SRM Institute of Science and Technology, Kattankulathur Campus,
Chennai – 603203, Tamil Nadu, India.
Abstract:- The aim of this research is to develop a model to available to perform machine learning tasks. Examples of
predict the price of a mobile when the specifications of a some frequently used algorithms include Naïve Bayes, K-NN,
mobile are given and to find the ML algorithm that etc. Feature selection algorithms can be used to select and
predicts the price most accurately. The usage of archival extract only the best parameters to train a model to optimize
data to accurately forecast forthcoming instances is the the accuracy and lessen the computational time of the model.
essence of Predictive Analytics. One of the ways Predictive Any of these methods can be used to perform the task of
Analytics can be performed is by using Machine Learning. predicting the price of a product depending on the type of data
Predictive Machine Learning works by taking in data as available to train the model.
input to develop and train a prediction model and the
trained model is used to predict the outcome of future data Nowadays, a cellphone is an essential accessory of a
instances. Supervised Machine Learning algorithms make person. It is the fastest evolving and moving product in the
use of data that contains a pre-defined class label, which is technology market space. New mobiles with updated versions
the attribute that needs to be predicted. The class label is and new features are introduced into the market at a rapid pace.
the price of a mobile in our case. The Mobile Price Class Thousands of mobiles are sold each day. In such a fast-paced
dataset sourced from the Kaggle data science community and volatile market, a mobile company needs to set optimal
website (https://www.kaggle.com/iabhishekofficial/mobile- prices to compete with its rivals. The first step in fixing a price
price-classification) that categorizes mobiles into price is to estimate the price based on the features. The objective this
ranges was used to train the prediction model. Python is research is to develop an ML model capable of estimating the
used due to its readily accessible ML libraries. Various price of a mobile phone based on its features. A potential buyer
classification algorithms were used to train the model to try can also make use of the model to estimate the price of a mobile
and find the algorithm that is able to predict the mobile by inputting just the features they require into the tool. The
price class most accurately. Metrics like accuracy score, same approach to create a prediction model can be used to
confusion matrix, etc. are were used to evaluate the trained develop a price estimation model for most products that have
model to determine the algorithm most suitable among the similar independent variable parameters. The price of a mobile
ones used. is dependent on many features for example, the processor,
battery capacity, camera quality, display size and thickness,
Keywords:- Machine Learning, Predictive Analytics, etc. These features can be used to classify phones into various
Supervised Machine Learning, Python. categories like entry-level, mid-range, flagship, premium, etc.
Supervised ML algorithms are used in this paper as the dataset
used has a definitive class label for price range.
I. INTRODUCTION
II. RESEARCH METHODOLOGY
Price is the most important component in the marketing
of any product and is often the definitive factor in its sale to a The research was carried out in Google Colab’s Python
consumer. In a constantly evolving and volatile market, the kernel. The general workflow diagram of supervised ML tasks
price is often the factor that makes or breaks a product. Setting is as follows:
an optimal price before the release of a product is imperative
for any company. A tool that gives the estimated price of a
product after weighing in the features it provides can come in
handy and can help the company in making an informed
decision while setting the market price for a product. Such a
tool can also be used by a consumer to get an estimated price
based on the features they are looking for in the product.
Machine learning algorithms can perform various tasks The dataset is portioned into two – train for training the
which are to be chosen with respect to the data to be worked model and test for its evaluation. The computer tries to
with and the motive of the task. Various tools and languages comprehend the logic behind the pricing of a mobile based on
like Python, MATLAB, Java, WEKA, Cygwin, Octave, etc are its features and uses it to forecast future instances as correctly
as possible.
IJISRT22JAN380 www.ijisrt.com 248
Volume 7, Issue 1, January – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
III. UNDERSTANDING THE DATASET IV. TRAINING THE PREDICTION MODEL
The Mobile Price Class dataset sourced from the Kaggle The first step in creating a model is to extract the required
data science community website features for training from the dataset and assigning the
(https://www.kaggle.com/iabhishekofficial/mobile-price- parameter that is to be the class label.
classification) that categorizes mobiles into price ranges was
used to train the prediction model.
The dataset contains 21 attributes in total – 20 features
and a class label which is the price range. The features include
battery capacity, RAM, weight, camera pixels, etc. The class In this code snippet, the first 20 attributes are being
label is the price range. It has 4 kinds of values – 0,1,2 and 3 extracted to serve as the training parameters and the final
which are of ordinal data type representing the increasing attribute (price_range) is used as the class label.
degree of price. Higher the value, higher is the price range the
mobile falls under. These 4 values can be interpreted as
economical, mid-range, flagship and premium.
So, despite price traditionally being a numeric problem,
the type of ML is classification (not regression) since there are The data is then portioned into two for the purpose of
discrete values in the class label. This is advantageous when training the model and testing it. A test size of 0.2 implies that
using algorithms like Naïve Bayes and Decision Tree as they 80% of the data is assigned to train the prediction model and
normally don’t work well with numeric data. the rest is utilized to measure the quality of the developed
model.
The dataset contains 2000 records in total.
Decision Tree was used here to train the prediction
model.
This is the numerical breakdown of the dataset:
Here, LDA was enforced to train the model.
IJISRT22JAN380 www.ijisrt.com 249
Volume 7, Issue 1, January – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Naïve Bayes algorithm was applied here.
KNN was used to train the model here. The certainty of LDA is found to be in the ballpark of 95%.
Random Forest was used to make the prediction model
here. All these algorithms are evaluated using various metrics
to find the algorithm most suited for the problem.
V. RESULTS AND DISCUSSION Naïve Bayes returned the correct classes with a veracity of
only 52.25% as Naïve Bayes is a poor classifier when
Metrics used to evaluate the algorithms in this paper are working with numeric data as input.
confusion matrix, classification report and accuracy score.
A confusion matrix has the total count of the accurately
grouped occurrences along its cross and the count of the
incorrectly classified instances in the rest of the matrix. We
have used 4 class values; so, the matrix generated is a 4*4
matrix.
A classification report gives the full report of the
classification with parameters like recall, precision, f1-score,
etc.
Accuracy score gives the accuracy of the trained model The efficiency of the model trained using the K-NN
after evaluating it using test data, for which we have sampled algorithm was found to be 92.75%.
20% of the dataset.
A veracity of 87% was achieved using Random Forest.
The algorithm that is found to be able to classify instances
the most accurately among the ones tested is LDA with an
accuracy of 95%, followed closely by KNN that was able to
predict instances with an accuracy of 92.75%. The Decision
Decision Tree was found to be able to correctly forecast Tree classifier and the Naïve Bayes classifier failed to forecast
the classes with a certainty of 75.75%. The reason for the the price range optimally.
average level of accuracy obtained is that the Decision Tree is
not suited for handling numeric data.
IJISRT22JAN380 www.ijisrt.com 250
Volume 7, Issue 1, January – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
VI. CONCLUSION
The model trained using LDA was found to predict
mobile price classes most accurately (95%). The accuracy of
the models can be improved by doing some data preprocessing
steps like normalization and standardization. Feature selection
and extraction algorithms can be used to remove unsuitable
and duplicative features to get better results. The same
procedure used in this paper can be applied to predict the prices
of other products like cars, bikes, houses, etc. using the
archival data containing features like cost, specifications, etc.
This would help organizations and consumers alike to make
more educated decisions when it comes to price.
REFERENCES
[1]. Mustafa Çetın, Yunus Koç, “Mobile Phone Price Class
Prediction Using Different Classification Algorithms
with Feature Selection and Parameter Optimization”,
IEEE, 2021, doi: 10.1109/ISMSIT52890.2021.9604550.
[2]. Muhammad Asim, Zafar Khan, “Mobile Price Class
prediction using Machine Learning Techniques”,
International Journal of Computer Applications (0975 –
8887) Volume 179 – No.29, March 2018, doi:
10.5120/ijca2018916555.
[3]. P. Arora, S. Srivastava and B. Garg, “MOBILE PRICE
PREDICTION USING WEKA”, 2020.
[4]. P. Durganjali and M.V. Pujitha, "House Resale Price
Prediction Using Classification Algorithms", 2019
International Conference on Smart Structures and
Systems (ICSSS), pp. 1-4, 2019.
[5]. D. Banerjee and S. Dutta, "Predicting the housing price
direction using machine learning techniques", 2017 IEEE
International Conference on Power Control Signals and
Instrumentation Engineering (ICPCSI), pp. 2998-3000,
2017.
[6]. Sameerchand Pudaruth . “Predicting the Price of Used
Cars using Machine Learning Techniques”, International
Journal of Information & Computation Technology.
ISSN 0974-2239 Volume 4, Number 7 (2014), pp. 753-
764.
[7]. Kanwal Noor and Sadaqat Jan, “Vehicle Price Prediction
System using Machine Learning Techniques”,
International Journal of Computer Applications (0975 –
8887) Volume 167 – No.9, June 2017.
[8]. R. Gareta, L.M. Romeo and A. Gil, "Forecasting of
electricity prices with neural networks", Energy
Conversion and Management, vol. 47, pp. 1770-1778,
2006.
[9]. https://www.kaggle.com/iabhishekofficial/mobile-price-
classification
[10]. https://www.gsmarena.com/
