Article
Machine Learning-Based Demand Forecasting for Short- and Long-Term Prediction
Demand forecasting plays a crucial role in effective decision-making across industries such as retail, supply chain, and inventory management. This work presents a data-driven machine learning framework for predicting short-term and long-term product demand using multiple algorithms. The system is designed to compare the performance of traditional and advanced models, including Random Forest, Gradient Boosting, LSTM, and XGBoost. Initially, the dataset is pre-processed through normalization, shuffling, and splitting into training and testing sets to ensure reliable evaluation. Each model is trained on 80% of the data and tested on the remaining 20% to measure performance. The evaluation is conducted using metrics such as R² score for accuracy and Root Mean Square Error (RMSE) for prediction error analysis. Experimental results show that Random Forest and Gradient Boosting provide moderate accuracy, while LSTM captures temporal patterns more effectively. However, the XGBoost model outperforms all other methods by achieving the highest accuracy and lowest error rate. The system also includes a user interface that enables real-time demand prediction for selected products. The results demonstrate that advanced ensemble techniques can significantly enhance forecasting performance. This approach provides a reliable and efficient solution for demand prediction in real-world applications. The proposed system can support better planning, reduce uncertainty, and improve operational efficiency.
Full Text Attachment
