Demand forecasting in the automotive spare parts industry is critical yet challenging due to the irregular and intermittent nature of demand patterns, the diversity of stock-keeping units (SKUs), and the economic impact of inventory mismanagement. This study explores the integration of spare parts classification techniques with machine learning models to enhance forecasting accuracy and operational efficiency, particularly in automotive sectors in developing countries. Traditional classification methods such as “ABC,” “FSN,” and “VED” are analyzed alongside advanced forecasting models, including Random Forest Regression and XGBoost regression, and compared with statistical methods including Simple Exponential Smoothing and Croston’s method. The study categorizes spare parts into fast moving, medium moving, slow moving, and non-moving items and evaluates forecasting methods based on performance metrics such as Mean Squared Error (MSE) and Root Mean Squared Error (RMSE). Results demonstrate the superior accuracy of machine learning models, especially Random Forest and XGBoost, in capturing diverse demand patterns and improving inventory management. The findings emphasize the necessity of tailored forecasting approaches for different spare parts categories, enabling businesses to optimize stock levels, reduce costs, and enhance customer satisfaction. This research underscores the potential of machine learning in addressing forecasting complexities, offering actionable insights for industry practitioners, and contributing to academic discourse on inventory and demand management.