Modeling online customer purchase intention behavior applying different feature engineering and classification techniques

In the evolution of digital technology, e-commerce sectors are gradually changing to realize customers’ demands and supply required things with low cost and due time. Recently, various machine learning techniques have been used to investigate different activities of customers and estimate different characteristics and requirements of customers. The goal of this work is to propose a machine-learning model that employs multiple data analytics and machine learning techniques to manipulate customer records and predict their buying intention more precisely. In this study, we collected an online shoppers’ purchasing intention dataset from a public data repository. Different feature transformation methods were employed in the primary dataset and generated its transformed datasets. Besides, we balanced the transformed datasets and detected outliers from them. Then, we applied different feature selection methods into primary and transformed-balanced datasets and again generated several feature subsets. Finally, various state-of-the-art classifiers were employed in primary, transformed, and all of their generated subsets. Then, different outcomes of the proposed model were analyzed and Random Forest was found as the stable classifier that produces more feasible results for any online shoppers’ buying instances. In this work, this classifier provided the best accuracy of 92.39% and f-score of 0.924 for the Z-Score and Gain Ratio transformed subset. In addition, it gave the highest AUROC of 0.975 for the Square Root and Information Gain subset. We also found Z-Score transformation and Information Gain more reliable methods to convert online shoppers’ customer intention dataset and get more feasible results from different classifiers.