Real-Time Cloud Analytics And Machine Learning For Demand Forecasting And Stock-Out Prevention

Authors: Dr. Pankaj Malik, Heer Wadhwani, Rigved Bhondve, Mehar Sahajpal, Liva Jain, Mahi Garg

Abstract: Stock-out events and inaccurate demand estimation are major challenges in modern inventory management, particularly in environments where sales and inventory conditions change rapidly. This paper proposes a Real-Time Cloud Analytics and Machine Learning framework for Demand Forecasting and Stock-Out Prevention that integrates cloud-based data processing, feature engineering, machine learning-based demand forecasting, inventory monitoring, and real-time stock-out risk assessment. The framework utilizes historical sales, inventory levels, price and promotion information, lead time, rolling demand, and temporal features to generate short-term demand forecasts. The predicted demand is combined with safety stock and reorder-point estimation to identify potential stock-out conditions and trigger replenishment decisions. A comparative evaluation of Random Forest, XGBoost, LightGBM, and the proposed cloud-ML framework is conducted using MAE, RMSE, and MAPE, together with stock-out and inventory-management measures. In the illustrative experimental results, the proposed framework achieves an MAE of 13.47, RMSE of 19.84, and MAPE of 7.12%, compared with 15.82, 22.76, and 8.35%, respectively, for LightGBM. The proposed framework also reduces stock-out events to 9, corresponding to a stock-out rate of 2.14%, while achieving a 97.86% service level and an average inventory level of 405 units. These results indicate the potential of integrating real-time cloud analytics with machine learning to improve demand forecasting, inventory visibility, and proactive stock-out prevention.

DOI: http://doi.org/10.5281/zenodo.22890767