Feature Selection and Adaptive Synthetic Sampling Approach for Optimizing Online Shopper Purchase Intent Prediction

Abstract

This study proposes a novel oncoming for optimizing online shopper purchase intent prediction using feature selection combined with Adaptive Synthetic Sampling (ADASYN). A supervised learning technique is applied to predict whether the customer visits ending with shopping or not based on the features. However, not all features are important to predict the classes. In addition, a suboptimal performance may occur due to the imbalanced class problem. Therefore, we propose Information Gain (IG) and Correlation (CORR) feature selection to select the most important features. ADASYN is additionally used to deal with the imbalanced class problem by adaptively generating new compositive samples for the minority class with considering density distribution. The proposed approach is run using Random Forest classifier. The experiments indicate that ADASYN successfully improves the classification results in the matter of accuracy, precision, recall, and F1-score. The use of feature selection combined with ADASYN has been compared to previous works, the results indicate that our proposed approach outperforms all. We additionally use a statistical test to show that our results are statistically significant. By these results, our proposed approach is proven for optimizing classification results.