PCA-Assisted K-Means Clustering of 31 Food Commodity Price Profiles in Indonesia

Abstract

Food price volatility exhibits heterogeneous characteristics across commodities, making uniform monitoring and stabilization strategies potentially ineffective. This study proposes a multidimensional clustering framework to identify homogeneous price profiles among 31 food commodities in Indonesia using Principal Component Analysis (PCA) and K-Means clustering. Daily commodity prices observed from September 2, 2019, to January 17, 2020, were transformed into statistical price characteristics, including mean, standard deviation, minimum, maximum, coefficient of variation, price changes, and volatility. The features were standardized using Z-score normalization and subsequently transformed using PCA to reduce redundancy while preserving the majority of information. The first ten principal components retained 97.39% of the total variance. Candidate cluster solutions were evaluated using the Elbow Method, Silhouette Score, Davies–Bouldin Index, and Calinski–Harabasz Index, resulting in an optimal three-cluster solution based on the combined validation criteria. The resulting clusters represent distinct multidimensional price profiles characterized by differences in price level, dispersion, price variation, and volatility. The proposed framework contributes a data-driven approach for grouping food commodities according to their joint price characteristics rather than considering price level alone. The resulting profiles can support differentiated monitoring priorities and more targeted food price stabilization policies.