A Mobile-Based Dynamic Approach to Comparative Study of Some Classification and Regression Techniques

Abstract

With the ever-growing complex world of Machine Learning algorithms and data processing techniques, the entry requirements for beginners are steadily rising. So, to allow a much lower entry ceiling into the world of Machine Learning, an Android application is proposed in this study that informs the user about the performance of a few particular supervised algorithms for any given data set. This study informs on three main aspects, namely an Android application that can be developed on flutter as the front end, an API that handles the upload of Comma Separated Values (CSV) file, processing of data, and sending the results back to the front end, and the Machine Learning pipeline that automates the data preprocessing, model creation, and model evaluation for the classification task. The data preprocessing includes data preparation with SMOTE, missing values filling, Scaling and Transformation of data, Feature Engineering, and Feature Selection. To build models, libraries such as Scikit-Learn, XGBoost, and LightGBM are used. Finally, when the user uses the application, they only need to upload a CSV file with the label of the data set renamed to “Target.” As soon as the file is uploaded, the server starts the automation process. Once the results are formed, the view of the application changes to a new screen where the results are displayed as a table.