Article Published In Vol.7 (July-Aug 2019)

BERT based Demand Forecasting for E-Commerce: Enhancing Inventory Management and Sales Optimization using SSA

Author : Himabindu Chetlapalli and Vinayagam.S

Accurate demand forecasting is essential for e-commerce platforms to optimize inventory management and enhance sales performance. Traditional forecasting models struggle with dynamic consumer behavior and seasonal demand fluctuations, leading to suboptimal stock levels and revenue loss. In this study, we propose a BERT-based Demand Forecasting Model integrated with the Squirrel Search Algorithm (SSA) to improve predictive accuracy. BERT efficiently captures contextual dependencies in sales data, while SSA optimizes hyperparameters for enhanced forecasting precision. Our model is evaluated on the Store Item Demand Forecasting dataset from Kaggle, benchmarked against ARIMA, LSTM, and Transformer-based models. The proposed BERT-SSA framework achieves a Mean Absolute Error (MAE) of 2.89, Root Mean Square Error (RMSE) of 4.35, and Mean Absolute Percentage Error (MAPE) of 1.92%, surpassing traditional models by 26.3% in MAE, 21.5% in RMSE, and 23.8% in MAPE. These improvements result in better demand stability across different product categories, reducing stockouts and overstocking risks. The experimental results validate that BERT-SSA effectively refines demand forecasting, leading to data-driven decision-making in inventory management. This study offers a scalable, adaptive AI-based forecasting framework that enhances supply chain efficiency and sales optimization for e-commerce businesses, empowering retailers with more accurate demand predictions and improved operational efficiency.

Keywords: BERT, Demand Forecasting, Squirrel Search Algorithm (SSA), E-Commerce, Time Series Prediction