<![CDATA[Computer Numerical Control (CNC) machining is a subtractive manufacturing technique that removes layers of material from a blank or workpiece to create a specific product. With increasing global competition, minimizing downtime during production is essential to maximize machine availability and productivity. This study investigates the application of machine learning models, specifically Extreme Gradient Boosting (XGBoost) and Random Forest (RF), to forecast CNC machine downtime from multiple failure sources. The study uses data collected from 16 CNC machines at Company A in Malaysia over an extended period. The data contain key variables for each downtime event, such as machine ID, failure type, start date/time, end date/time, and downtime duration in minutes. Failure types are categorized into several groups, including mechanical, electrical, and tool malfunctions. After hyperparameter tuning, the XGBoost model outperformed the RF model, achieving a Mean Squared Error (MSE) of 0.4017, Root MSE (RMSE) of 0.634, and Mean Absolute Error (MAE) of 0.470 on the test set, while the RF model yielded higher errors, with an MSE of 1.2654, RMSE of 1.125, and MAE of 0.943. These results demonstrate the superiority of the XGBoost model over RF in predicting future CNC downtime, as indicated by its lower prediction errors. Future work should focus on refining the model with larger, more diverse datasets and exploring its integration into AI-based decision support systems to enhance machine availability and operational efficiency.]]>
