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 vast amounts of transactional and behavioral data, creating new opportunities for advanced predictive analytics in retail environments. Sales prediction has become a critical application, enabling businesses to forecast demand, optimize inventory management, and improve supply chain efficiency while reducing operational costs and enhancing customer satisfaction. Traditional forecasting techniques such as ARIMA and regression models have been widely used; however, they often struggle to capture the nonlinear relationships and complex patterns present in modern e-commerce datasets. Recent advancements in machine learning and deep learning, including random forests, support vector machines, convolutional neural networks (CNNs), and long short-term memory (LSTM) networks, have significantly improved prediction accuracy by effectively modeling temporal and spatial patterns. Hybrid CNN-LSTM architectures further enhance performance by combining feature extraction and sequence learning capabilities. Additionally, Siamese neural networks enable the analysis of heterogeneous data sources by learning similarity relationships across inputs, making them particularly useful for new product demand prediction. Optimization techniques such as the Giant Trevally Optimizer further enhance model performance, leading to more accurate and scalable e-commerce forecasting systems.