COMPARATIVE ANALYSIS OF BCBIMAX AND PLAID BICLUSTERING ALGORITHM FOR PATTERN RECOGNITION IN INDONESIA FOOD SECURITY

Abstract

Biclustering is an unsupervised learning algorithm that simultaneously groups rows and columns in a data matrix. Unlike conventional clustering, which evaluates objects across all variables independently, biclustering identifies subsets of objects and variables that share similar patterns—revealing localized structures within complex datasets. This study applies the BCBimax and Plaid algorithms to examine food security patterns across 34 Indonesian provinces. The indicators cover three key dimensions: availability, accessibility, and utilization of food. The algorithms are evaluated using the Jaccard Index, Mean Squared Residue (MSR), and the number of provinces effectively clustered. Results show that BCBimax, using a binarization threshold based on the median value, generates eight biclusters covering 58.8% of provinces. Meanwhile, the Plaid algorithm, applying constant column model parameters, produces six biclusters with 55.88% coverage, including overlapping memberships. Overall, BCBimax demonstrates superior performance, as indicated by a lower average MSR value (0.035) compared to Plaid (0.209). The Jaccard Index similarity score of 14.61% suggests that the biclusters formed by each method are significantly distinct. Both approaches indicate that the majority of Indonesian regions exhibit low to moderate food security characteristics.