Pattern Recognition of Food Security in Indonesia Using Biclustering Plaid Model

Abstract

Listen

Translate article

Biclustering come in various algorithms, selecting the most suitable biclustering algorithm can be a challenging task. The performance of algorithms can vary significantly depending on the specific data characteristics. The Plaid model is one of popular biclustering algorithms, has gained recognition for its efficiency and versatility across various applications, including food security. Indonesia deals with complex food security challenges. The nation's unique geographic and socioeconomic diversity demands region-specific food security solutions. Identifying province-specific food security patterns is crucial for effective policymaking and resource allocation, ultimately promoting food sufficiency and stability at the regional level. This study assesses the performance of the Plaid model in identifying food security patterns at the provincial level in Indonesia. To optimize biclusters, we explore various parameter tuning scenarios (the choice of model, the number of layers, and the threshold value for row and column releases). The selection criteria are based on the change ratio of the initial matrix's mean square residue to the mean square residue of the Plaid model, the average mean square residue, and the number of biclusters. The constant column model was selected with a mean square residue change ratio of 0.52, an average mean square plaid model residue of 4.81, and it generates 6 overlapping biclusters. The results show each bicluster has unique characteristics. Notably, Bicluster 1 that consist of 2 provinces, exhibits the lowest food security levels, marked by variables X1, X2, X4, and X7. Furthermore, the variables X1, X4, and X7 consistently appear across several biclusters. This highlights the importance of prioritizing these three variables to improve the food security status of the regions.