Hafsa E

Inventory Management Optimization through Deep Learning: A Hybrid LSTM-Q-Learning Approach

Ibn tofaïl university - Kénitra • Part-time

Oct 2025 - Jan 2026 • 3 mos

Optimal inventory management remains a major challenge for companies facing increasing demand volatility and the complexity of modern supply chains. Traditional methods such as ARIMA models or classical replenishment policies (EOQ, (s,S)) struggle to capture the non-stationarity and complex temporal dependencies inherent in contemporary demand series. This article proposes a hybrid approach combining LSTM (Long Short-Term Memory) networks for demand forecasting and reinforcement learning via Q-Learning for order decision optimization. Our methodology formalizes the problem as a Markov decision process where the state integrates current inventory, LSTM forecasts, and contextual variables. Experimental results on a real sales dataset show a 23% reduction in total costs compared to baseline methods, a 35% decrease in stockouts, and an 18% improvement in customer service level. This research demonstrates that integrating deep learning techniques into inventory management systems enables superior performance while dynamically adapting to demand variations.