<![CDATA[Low Birth Weight (LBW) remains a major maternal and child health challenge associated with elevated risks of neonatal mortality and stunting. Conventional one-way clustering algorithms partition health centers or risk indicators separately, failing to identify local interactions across specific subsets of variables simultaneously. This study conducts a comparative evaluation of three normalization schemes within the Spectral Biclustering algorithm—Independent Rescaling of Rows and Columns (IRRC), Bistochastization, and Log-Interactions—to identify optimal bicluster structures on 2024 LBW data across 95 public health centers in Gorontalo Province. Preprocessing procedures addressed missing observations using mean imputation prior to variable standardization via Z-scores. The parameter tuning framework systematically evaluated 105 parameter combinations across variations of singular vectors (E = 2 to 8) and intra-bicluster variance thresholds (withinvar = 1 to 5), yielding 17 candidate models that satisfied stability criteria. Model validation using the Average Squared Residue (ASR) demonstrates that Bistochastization achieves optimal bicluster homogeneity, generating eight biclusters with the lowest ASR value of 0.1755535 at configuration E = 6 and withinvar = 3. In comparison, Log-Interactions and IRRC achieved higher ASR values of 0.3552414 (eight biclusters) and 0.3623521 (six biclusters), respectively. Structural bicluster simplification reveals distinct spatial patterns, grouping maternal health service disparities and maternal physiological risk profiles across specific public health centers. These empirical findings provide an analytical foundation for targeted maternal health interventions across regional health centers.]]>
Copyrights © 2026
