Deep Learning Approach for Inventory Stock Prediction Using LSTM Networks

Abstract

Accurate inventory demand forecasting is essential for maintaining optimal stock levels, minimizing operational costs, and improving supply chain efficiency in modern retail and warehousing environments. However, conventional statistical and machine learning approaches often fail to effectively capture nonlinear demand fluctuations, seasonal variations, and long-term temporal dependencies present in real-world inventory datasets. Moreover, the absence of real-time inventory monitoring limits the responsiveness and reliability of existing forecasting systems. To address these challenges, this study proposes an IoT-enabled deep learning framework for intelligent inventory demand forecasting using Long Short-Term Memory (LSTM) networks. The proposed system integrates IoT devices, QR-code-based inventory tracking, cloud-based data management, and LSTM-based predictive analytics to enable real-time stock monitoring and accurate demand predictions. A comprehensive preprocessing pipeline, including data cleaning, normalization, categorical encoding, and sequence generation is implemented to improve model learning efficiency and forecasting stability. The proposed framework was performed using MAE, RMSE, MAPE, and $\mathbf{R}^{\mathbf{2}}$ metrics, and the proposed framework achieved an MAE of 0.0798, RMSE of 0.0876, MAPE of 4.12%, and an $\mathbf{R}^{\mathbf{2}}$ score of 0.96, outperforming conventional machine learning and deep learning forecasting models including Random Forest, XGBoost, RNN, GRU, and CNN-LSTM. The experimental results demonstrate that the proposed framework effectively captures long-term temporal dependencies and nonlinear demand patterns, resulting in improved forecasting accuracy, stability, and intelligent inventory optimization. The proposed system offers a scalable and intelligent solution for smart retail management, warehouse automation, and digital supply chain optimization.