Mobile Phone Price Class Prediction Using Different Classification Algorithms with Feature Selection and Parameter Optimization

Abstract

Machine Learning (ML) algorithms are used in many fields such as finance, education, industry, medicine and e-commerce. ML algorithms show performance differences depending on the dataset and processing steps. Choosing the right algorithm, preprocessing and postprocessing methods has great importance to achieve good results. In this paper, Random Forest Classifier, Logistic Regression Classifier, Decision Tree Classifier, Linear Discriminant Analysis, K-Nearest Neighbor Classifier and SVC methods are compared to predict mobile phone price class. The "Mobile Price Classification" dataset which is taken from Kaggle.com is used to evaluate methods. Firstly, all values at the dataset are checked to verify that there are no missing values. After that, scaling is applied to dataset in order to obtain more relevant data for ML algorithms. Then, feature selection methods which reduce the computational cost by reducing the number of inputs are performed to get meaningful features. Finally, the parameters of classification algorithms are tuned to improve the system accuracy. According to obtained results, it is seen that ANOVA f-test feature selection method is more convenient for this dataset. It gives satisfying accuracy with a minimum number of features. It is also seen that the SVC classifier has the highest test accuracy compared to other models.