Enhancing Accuracy in Recommender Systems with a Hybrid Deep Learning Approach for Web Usage Mining

The rapid expansion of e-commerce platforms has created an urgent demand for intelligent recommender systems capable of providing personalized and context-aware product recommendations. While traditional recommendation models offer some effectiveness, they often face challenges such as the cold start problem, data sparsity, and the inability to capture sequential user behavior. This paper presents a hybrid deep learning-based recommendation architecture that combines a Deep Feedforward Neural Network (DFNN) with a Long Short-Term Memory (LSTM) network to overcome these limitations. The proposed hybrid model was evaluated against standalone DFNN and LSTM models using a real-world e-commerce clickstream dataset. The results demonstrate the hybrid model's superior performance, achieving a training accuracy of 99.06% and validation accuracy of 98.98%. Additionally, the model excels in classification performance, with precision, recall, and F1-scores approaching 1.00 across critical user actions such as add_to_cart, purchase, search, and view.