Deciphering Key Features in Online Buying Session Prediction Using Logistic Regression and Naive Bayes

This paper explores the key factors influencing online buying session prediction using nomogram analysis with Logistic Regression and Naive Bayes models. We analyzed a dataset containing various attributes of online shoppers. Our results underscore that the Logistic Regression has superior capability for both True and False targets in terms of accuracy (0.883) and AUC (0.897). While Naïve Bayes performs better for the True target with an F1Score of 0.555, Logistic Regression demonstrates superior performance for the False target with an F1 Score of 0.934. Nomogram analysis revealed that for Naive Bayes, page values, exit rates, and bounce rates are the most significant factors, while for Logistic Regression, page values and product-related features (ProductRelated and ProductRelated_Duration) are the most significant. Visualizing these key factors through nomograms helps interpret their relative importance to support the accuracy and interpretability of machine learning predictions.