This study addresses the challenge of forecasting unplanned repair demand for military helicopter engine components, which is responsible for depot-level maintenance of Turkish Land Aviation aircraft. Current demand forecasting relies heavily on expert intuition and simple historical averages, creating a person-dependent planning process that cannot quantify forecast errors. Using Ishikawa cause-and-effect analysis, the root causes of uncertainty were identified across six categories; Pareto analysis revealed that Material & Supply and Forecasting Method deficiencies account for approximately 50% of the problem burden. To address these, a Decision Support System (DSS) was developed, incorporating an automated data cleaning pipeline, KPI computation, and ten forecasting models: Seasonal ARIMA, Exponential Smoothing, Random Forest, XGBoost, LightGBM, CatBoost, Prophet, Croston, SBA, TSB. Walk-forward validation shows that gradient boosting models (especially LightGBM) achieve the lowest MAE (≈15.5), reducing forecast error by 24.3% compared to the existing operational benchmark (Seasonal ARIMA) used at the depot. Lagged demand features proved most influential via permutation importance. The DSS is deployed as an interactive dashboard, standardizing the planning process and enabling proactive procurement decisions.

Keywords

Military Helicopter Maintenance, Unplanned Repair Demand Forecasting, Decision Support System, Time Series, Machine Learning.