EEMD-LSTM Framework for Predictive Analytics in E-commerce Platforms

Abstract

E-commerce platforms face dynamic customer demands, highly volatile sales patterns, and increasing competition, making accurate forecasting critical to maintaining competitive advantage. Traditional time series forecasting methods struggle to adapt to nonlinear and no stationary data patterns commonly observed in e-commerce. In this paper, we propose an advanced hybrid framework integrating Ensemble Empirical Mode Decomposition (EEMD) with Long Short-Term Memory (LSTM) networks for predictive analytics in e-commerce environments. EEMD decomposes complex sales signals into simpler Intrinsic Mode Functions (IMFs), isolating noise and revealing inherent patterns. These IMFs are individually modeled using LSTM networks, leveraging their capacity to capture long-term dependencies and nonlinear structures. Extensive experiments were conducted using real-world e-commerce datasets, comparing the EEMD-LSTM framework with benchmark models such as ARIMA, Prophet, and standalone LSTM. Our proposed method outperformed traditional and deep learning baselines in terms of Root Mean Square Error (RMSE) and Mean Absolute Error (MAE), demonstrating the robustness and effectiveness of the hybrid approach. This research highlights the potential of signal decomposition combined with deep learning for accurate and adaptive predictive analytics in rapidly evolving e-commerce markets.