AN INTELLIGENT CONSUMER BEHAVIOUR PREDICTION SYSTEM USING HYBRID MACHINE LEARNING ALGORITHMS

Predicting whether a website visitor will make a purchase is one of the most commercially valuable problems in online marketing, yet it is difficult because conversion is rare and visitor behaviour is noisy. This paper presents an intelligent consumer behaviour prediction system that combines unsupervised segmentation with a stacked ensemble of Random Forest, XGBoost and LightGBM learners, joined by a logistic-regression meta-learner. The system was trained and evaluated on the public Online Shoppers Purchasing Intention dataset, which holds 12,330 browsing sessions of which 15.5% ended in a purchase. After cleaning, encoding and engineering three aggregate behavioural features, an 80/20 stratified split was used for evaluation. The hybrid model reached an AUC of 0.929, the highest of all nine configurations tested, and, after a decision threshold was tuned on training data only, an F1-score of 0.678 with a recall of 0.715 on the held-out sessions. A baseline ablation shows that most of the gain comes from stacking heterogeneous learners, while the clustering step mainly contributes interpretability by separating a segment with a 29.4% conversion rate from one with 1.8%. Page value, exit rate and time spent emerged as the strongest predictors. The paper discusses managerial uses of the system and the limits of a purely behavioural data source, and proposes the inclusion of psychological signals such as fear of missing out, urge for instant gratification and social validation in later versions