KLASTERISASI POLA PERMINTAAN PADA UMKM MENGGUNAKAN DATA TRANSFORMATION DAN K-MEANS CLUSTERING

Abstract

The consumer electronics retail industry is characterized by short product life cycles and intermittent demand patterns at the stock keeping unit (SKU) level. This study performs demand pattern segmentation of SKUs at an electronics retail store using the Average Demand Interval (ADI) and Squared Coefficient of Variation (CV²) features extracted from three years of transaction data. Both features were transformed using log1p and Z-score standardization to address the skewed data distribution, and then clustered using the K-Means algorithm with candidate values of K ranging from 2 to 4. The optimal number of clusters was selected based on the Silhouette Score and Davies-Bouldin Index (DBI), while segment labeling was performed using empirical median thresholds derived from the research data, rather than the standard Syntetos-Boylan-Croston cut-off values. Results on 1,342 valid SKUs show that K=2 is the optimal number of clusters (Silhouette Score 0.6380; DBI 0.7034), which were interpreted as the "routine-stable demand" segment (1,055 SKUs; 78.61%) and the "routine but fluctuating" segment (287 SKUs; 21.39%), with the separation being more dominated by ADI values than by CV². These findings indicate that the demand structure is more accurately represented by two transaction-frequency-based groups, and can serve as a basis for inventory management policies tailored to the characteristics of each segment.