To reduce inventory costs for aviation enterprises, an end-to-end multi-period inventory optimization model based on machine learning was established, taking into account historical demand data and feature data for maintenance spare parts. The objective was to minimize total inventory costs, and dynamic programming is employed to determine the historical optimal replenishment decisions. Building on this foundation, an end-to-end inventory optimization model was constructed, where TimeXer served as the demand forecasting module and XGBoost functions as the inventory decision module. The integration of these two components enhanced the scientific basis of replenishment decisions. An example from a large aviation maintenance enterprise demonstrates that this end-to-end model effectively reduces inventory costs and exhibits high practical application value and feasibility for various types of consumable aviation spare parts.