Comparative Evaluation of K-Means and Hierarchical Clustering for Multi-Dimensional Criticality Classification and Differentiated Stocking of Aerospace Spare Parts

Aerospace maintenance organizations manage hundreds of thousands of unique spare parts across multi-echelon supply networks, where part shortages directly impair aircraft readiness and excess inventory ties up substantial resources. Traditional ABC and ABC-XYZ classification methods rely on one or two criteria and fail to capture the multi-dimensional criticality that distinguishes aerospace spare parts from general inventory. This paper presents a comparative evaluation of K-Means and agglomerative hierarchical clustering for multi-dimensional criticality classification of aerospace spare parts using six features: average monthly demand, demand variability, unit cost, Aircraft-on-Ground criticality, supply source count, and replenishment lead time. Clustering quality is assessed using the Silhouette Score and the Davies-Bouldin Index across cluster counts from 2 to 6. Experiments based on 5,000 Royal Air Force spare parts demand records, combined with criticality and supply-risk attributes literature-calibrated from published aerospace parameter distributions, indicate that K-Means at four clusters provides the strongest overall internal validity trade-off, including the highest Silhouette Score of 0.437 and the lowest Davies-Bouldin index of 0.891 among the reported configurations. Under the resulting scenario-based policy approximation, differentiated stocking rules mapped to the criticality clusters are estimated to reduce total safety stock investment by 12.3% while yielding an expected weighted-average service level above 96%, compared to a uniform 95% baseline. These findings illustrate a data-driven and managerially interpretable framework for prioritizing inventory resources in multi-echelon aerospace supply networks under scenario-based criticality assumptions.