Machine Learning Models for Predicting Online Shoppers Purchase Intention

Abstract—The customer purchase intention is a task that isuseful in e-commerce websites in predicting successful customerpurchase intention in order to enhance marketing strategiesand improve sales conversion. In the current paper, the onlineshoppers intention dataset will be used to present a machinelearning model that predicts the behavior of online shoppers (whether to buy or not to buy). The data was initially pre-treated, i.e. the class imbalance and the outliers were eliminated through the Interquartile Range (IQR) approach to enhancethe quality of the data and the performance of the model. Thefeature importance scores through random forest were used toselect and maintain only the most relevant behavioral attributes.This step minimized the unnecessary features and enhanced theperformance of computing. A number of classification algorithmswere trained and tested including Logistic Regression, RandomForest, Gradient Boosting, and a Stacking ensemble model.Experiment outcome indicates that ensemble learning performswell than individual models and attains accuracy of about 93.84%after preprocessing and feature optimization. The suggestedmethod illustrates that a high data cleaning and feature selectionand ensemble can play a big role in improving prediction in onlinepurchase intention analysis.