Analysis of Classification Algorithms for the Prediction of Purchase Intention in Electronic Commerce

Abstract

The analysis focuses on data from an e-commerce retailer to identify the classification techniques for obtaining the best performances of validated classification algorithms. It is essential to highlight that this type of prediction is used in the commerce sector since it is a sector in constant development and allows to identify the consumer’s behaviour to variables such as average time on the website, transactions completed, pages rated, among others. The research focuses on Business to Consumer commerce, collecting data from the purchase sessions for 12 months. This allowed the comparison of the Naive Bayes model of Gaussian type, Random Forest and Extra Trees Classifier, which are finally validated against the results of Dummy classifier, highlighting the efficiency of Extra Trees Classifier. This type of study and the selection of the variables provide a higher correlation with the Revenue attribute, which is the one used as a prediction class. The efficiency of Extra Trees Classifier stands out in this type of study, together with the selection of the variables that provide the highest correlation with the Revenue attribute, which is the one used as the prediction class.