Data-Driven Systems: Engineering Foundations for Scalable Intelligence

Abstract

The rapid expansion of digital commerce platforms has intensified the demand for scalable and intelligent recommendation systems capable of adapting to dynamic user behavior. Traditional approaches based on conventional machine learning (ML) and basic collaborative filtering (CF) often exhibit limitations in capturing sequential interaction patterns, handling sparse, large-scale data, and adapting to temporal variations in user preferences. To address these challenges, develop a scalable data-driven model that enhances recommendation accuracy through advanced deep learning (DL) mechanisms. The novelty lies in integrating temporal attention with a gated recurrent architecture and optimizing its performance using a bio-inspired optimization strategy. The E-Commerce User Behavior Dataset consists of 12,000 records, containing user interaction events such as views, cart additions, and purchases. Preprocessing includes data cleaning and duplicate removal to ensure temporal consistency and data quality. Feature extraction is achieved through behavioral and sequential pattern representation using an autoencoder (AE) to learn latent user features, capturing user preferences and interaction dynamics. The proposed model employs Temporal Attention -based Gated Recurrent Unit integrated with the Giant Trevally Optimizer (TAGRU-GTO) to sequential user interactions and emphasizes significant behavioral transitions. The TAGRU component models temporal dependencies while the attention mechanism prioritizes influential events, and the optimization strategy fine-tunes hyperparameters to improve convergence efficiency. The model achieves a validation accuracy of 98.7%, a test accuracy of 98.6%, and a test RMSE of 0.74. Implementation is conducted using Python-based DL methods. Experimental outcomes demonstrate improved recommendation accuracy, scalability, and robustness compared to baseline methods.