ANALYZING THE EFFICACY OF ML APPROACHES IN INVENTORY FORECASTING: AN IN-DEPTH COMPARISON OF LSTM, PROPHET, AND ENSEMBLE APPROACHES
Patel Aarsh Miteshkumar1, Patel Bhavya Manishkumar2, Dhruv Pandey3, Dr. Renuka Devi R4
1,2,3UG Student, 4Associate Professor,
School of Computer Science Engineering, Vellore Institute of Technology, Chennai, Tamil Nadu, India
ABSTRACT—In retail business, stock levels optimization is one of the significant aspects, where changes can have a great impact on profitability and customer satisfaction. for this purpose, this paper introduces the RetailRevolutionizer: A Machine Learning- based Inventory Demand Forecasting tool designed for comparison of three advanced models namely LSTM, Prophet, and Prophet + SARIMA, by evaluating each model’s performance in terms of forecasting accuracy, RetailRevolutionizer provides insights into their strengths and limitations. The LSTM captures the long-term dependencies effectively, while SARIMA deals well with seasonality. Prophet specifically specializes in trend decomposition and anomaly detection The overall tool implementation consists of data cleaning, trend analysis, study of sales distribution, and detecting anomalies. Through the trends for sales clustering and sales time-series decomposition, RetailRevolutionizer delivers actionable insights into sales variability and peak periods. It has proven that improvements in every model are appreciated by the augmentation in forecasting accuracy and avoided cases of overstocking and under-stocking. This ensures better inventory turns and increases connectivity with customers’ demand, making the overall operating efficiency better.
INDEX TERMS—Inventory Management, LSTM, SARIMA, Prophet, Predictive Analytics, Machine Learning