Forecasting Online Retail Sales with Empirical Mode Decomposition and Deep LSTM

Abstract

Forecasting online retail sales has become a critical task for e-commerce enterprises aiming to maintain competitive advantage, manage inventory efficiently, and optimize marketing strategies. Traditional forecasting models often fall short in capturing the non-linear and non-stationary nature of retail sales data, leading to suboptimal accuracy and strategic misjudgments. This paper proposes a hybrid forecasting model that leverages Empirical Mode Decomposition (EMD) for preprocessing the sales data and a Deep Long Short-Term Memory (LSTM) neural network for learning temporal dependencies. EMD serves to break down complex sales signals into Intrinsic Mode Functions (IMFs), enabling the model to extract meaningful patterns from volatile and noisy sales data. Our proposed model is evaluated using a real-world online retail dataset comprising transaction records over a multi-year period. Results demonstrate that the EMD-Deep LSTM model significantly outperforms traditional models such as ARIMA, standard LSTM, and Prophet in terms of RMSE, MAE, and MAPE. The research offers empirical evidence that the combination of EMD and Deep LSTM can serve as a powerful tool for sales forecasting in dynamic online retail environments.