A Survey of Methods and Architectures for E-Commerce Systems for Sales Prediction Using Triple Pseudo-Siamese Network with Giant Trevally Optimizer

Abstract

The rapid expansion of digital commerce platforms has generated large volumes of transactional and behavioral data, enabling advanced predictive analytics in retail environments. Sales prediction is a key application that supports demand forecasting, inventory optimization, and supply chain efficiency. Accurate forecasting helps businesses reduce operational costs, prevent stock shortages, and improve customer satisfaction. Traditional models such as autoregressive integrated moving average (ARIMA) and regression techniques have been widely used; however, they often fail to capture nonlinear patterns and complex relationships in modern e-commerce datasets. Recent advancements in machine learning and deep learning have significantly improved forecasting performance. Techniques such as random forests, support vector machines, convolutional neural networks (CNN), and long short-term memory (LSTM) networks effectively model large-scale and temporal data. Hybrid CNN-LSTM architectures further enhance prediction accuracy by combining spatial and temporal feature extraction. Additionally, Siamese neural networks enable learning from heterogeneous data sources by capturing relationships between products, customers, and transactions. Optimization methods like the Giant Trevally Optimizer (GTO) improve model convergence and parameter tuning. This review analyzes recent studies and highlights hybrid frameworks for scalable, accurate sales prediction systems.