A Multi-Model Approach to Supermarket Sales Forecasting System Using Machine Learning and Time Series

Supermarkets frequently encounter challenges in managing their inventory efficiently, often resulting in significant wastage due to overstocking or understocking. This research aims to enhance supermarket inventory management by implementing a robust sales prediction and forecasting system using advanced machine learning algorithms. The primary objectives of this study are to reduce inventory wastage, optimise stock levels, and increase profitability by accurately predicting product demand based on weather forecasts and economic factors. The proposed solution is highly beneficial for supermarket managers, providing actionable insights for inventory decisions.The system integrates ARIMA (AutoRegressive Integrated Moving Average) and multiple machine learning models, , Randomforest, and XGBoost (Extreme Gradient Boosting), to deliver precise sales forecasts. By combining these models through a stacking ensemble approach, the solution leverages the strengths of each model to enhance overall prediction accuracy. This paper presents the development, implementation, and evaluation of the solution, highlighting its potential to transform inventory management practices in the retail industry. The backend of the system is supported by a comprehensive database, ensuring seamless data processing and real-time updates. The web-based application offers an intuitive interface for users, enabling efficient inventory management. The results demonstrate significant improvements in sales forecasting accuracy, which can lead to reduced wastage and better inventory control in supermarkets.