Prediction Framework for E-Commerce Platform Sales Data Based on Informer: A Study on Furniture Sales on Amazon E-Commerce Platform

Abstract

Accurate sales forecasting is a critical requirement for e-commerce platforms, as it supports efficient inventory management, dynamic pricing, and targeted marketing strategies. Nonetheless, forecasting is inherently challenging due to the volatility of consumer demand, the impact of external factors such as economic indicators and weather conditions, and the occurrence of promotional events. To address these complexities, this study proposes a prediction framework based on the Informer model, a Transformer-based architecture specifically designed for long-sequence time series forecasting. The framework leverages weekly sales data of five furniture categories from the Amazon e-commerce platform spanning 2015 to 2020, combined with auxiliary features including price, discount rate, consumer price index (CPI), and weather-related variables. By integrating both temporal dependencies and exogenous influences, the model captures intricate patterns in sales dynamics. Experimental results demonstrate that the Informer model significantly outperforms traditional baselines, including LSTM, GRU, and Prophet, achieving a mean absolute error (MAE) of 37.3 and a root mean square error (RMSE) of 52.7. These findings highlight the model’s effectiveness in enhancing e-commerce sales forecasting and underscore the potential of advanced deep learning techniques to improve operational decision-making for digital retail platforms.