Abstract
<jats:p>Biclustering is a two-way clustering method that identifies local patterns simultaneously across rows and columns of a data matrix. However, missing values may alter data structures and affect biclustering results. Studies evaluating the interaction between imputation methods and biclustering algorithms remain limited. This study evaluates the performance of the Iterative Signature Algorithm (ISA) and Plaid Model following imputation using Hot Deck, K-Nearest Neighbor (KNN), and Expectation Maximization (EM). The novelty of this study lies in assessing how the interaction between imputation methods and biclustering algorithms affects bicluster recovery and quality. Missing values were generated under MCAR at 5% and 10% proportions with 100 repetitions. Bicluster quality was evaluated using Mean Squared Residue (MSR), Transposed Virtual Error (VEt), and Sub-Matrix Correlation Score (SCS), while bicluster consistency was assessed using the Jaccard Index (JI). ISA consistently achieved higher JI values, indicating better preservation of bicluster structures, whereas the Plaid Model produced lower MSR, VEt, and SCS values, indicating more homogeneous biclusters. KNN generally showed the most consistent performance across scenarios. These findings suggest that imputation methods and biclustering algorithms should be selected jointly according to the analytical objective to obtain reliable biclustering results from incomplete macroeconomic data.</jats:p>