The demand for aircraft parts is often difficult to forecast, leading to challenges in maintenance planning, inventory management, and overall operational efficiency. In this invited paper, we examine monthly demand forecasting for used aircraft parts in the nonintermittent domain (Smooth and Erratic), with CatBoost as the forecasting model. Our dataset contains more than 1,000 part-by-month demand records from January 2022 to June 2024, and we measure forecast accuracy using Mean Absolute Error (MAE). The best CatBoost model is a two-feature configuration that uses dispersion, expressed as Coefficient of Variation Squared (CV 2 ), and a sine encoding of the month of year, expressed as Month Sin. Against classical Croston-family baselines (Croston, SBA, and TSB), the CatBoost model reduces MAE by about 14%. From an operations standpoint, smaller under- and over-forecast errors support leaner safety stock levels, fewer rush shipments, and lower Aircraft on Ground (AOG) exposure.
Leevy et al. (Fri,) studied this question.