SMARTCART AI: PERSONALIZED ITEM CLASSIFICATION FOR ENHANCED E-COMMERCE EXPERIENCE

Abstract

In the dynamic landscape of personalized retail, over 85% of consumers express preference for tailored product recommendations, yet only 40% of current systems effectively adapt to evolving customer behavior patterns, highlighting a critical gap in personalization intelligence. Traditional classification systems often suffer from limited contextual understanding and lack robustness in handling complex customer-product interactions, leading to sub-optimal product targeting and recommendation accuracy. To address these challenges, this research presents a novel retail item classification framework aimed at predicting customer preference levels—Neutral, Preferred, and Trending—using an enriched feature set derived from real-world shopping metrics. The study utilizes a synthesized dataset containing diverse attributes such as price, user preference scores, popularity, ratings, discounts, seasonality, and stock availability. A meticulous preprocessing phase includes label encoding of categorical variables and exploratory data analysis (EDA), incorporating visual techniques like count plots for class distribution and heatmaps to reveal inter-feature correlations, ensuring a well-informed modeling foundation. Initially, a baseline Multi-Layer Perceptron (MLP) Classifier is implemented to benchmark performance. However, to improve precision and generalization, the study introduces a Gradient Boosting Classifier, which leverages ensemble learning and stage-wise optimization to capture intricate decision boundaries. This proposed model significantly outperforms the existing one across accuracy, precision, recall, and F1-score metrics, demonstrating its superior capability in customer preference classification. The results indicate that integrating feature importance-driven preprocessing with robust boosting algorithms provides a scalable and intelligent solution for real-time retail analytics