Enhancing E-commerce Supply Chains with Time-Series Forecasting Using Long Short-Term Memory (LSTM) Networks

E-commerce supply chains function in ever-changing settings, requiring precise demand forecasting to ensure efficient operations, reduce costs, and improve customer satisfaction. Traditional forecasting methods often fail to capture non-linear patterns and temporal dependencies inherent in time-series data, limiting their effectiveness. This paper proposes an ESCLSTM technique to enhance e-commerce supply chains (ESC) by leveraging Long ShortTerm Memory (LSTM) networks to analyse data sequentially and forecast. The ESCLSTM methodology begins with collecting historical sales and inventory data from an open-source ecommerce dataset. The data is preprocessed through missing value imputation, normalization, and time-series decomposition to prepare it for modelling. LSTM networks, designed to capture sequential dependencies, are appraised and trained using measures like Mean Absolute and Root Mean Square Error, with hyperparameters optimized through grid search. Key findings reveal that the LSTM model outperforms traditional approaches, achieving a 20% reduction in MAE and a 25% improvement in RMSE. The multivariate LSTM model demonstrates superior performance in capturing complex relationships between features, leading to more accurate predictions of demand trends. In conclusion, the study highlights the potential of LSTM networks to revolutionize demand forecasting in e-commerce supply chains, offering a robust and scalable solution for handling the complexities of modern logistics.

Keywords :

E-commerce supply chains, Demand forecasting, Time-series forecasting, Long ShortTerm Memory (LSTM) networks, Sequential data analysis.