Klasifikasi Kategori Indeks dmft pada Data Pemeriksaan Gigi Menggunakan Naïve Bayes dengan Seleksi Fitur Kendall's Tau

Abstract

Dental caries severity in primary dentition can be summarized as ordinal dmft index categories. This study evaluates Gaussian Naïve Bayes for classifying these categories and explores Kendall's Tau-based feature selection. Kendall's Tau was selected as a nonparametric rank-based filter because the outcome is ordinal and the method does not require normally distributed variables, although its application to nominal predictors encoded as integers remains exploratory. Of 3,046 dental-examination records, 1,285 were retained after cleaning; 1,761 records (57.81%) were excluded mainly because responses were incomplete or could not be mapped consistently. A fixed 80:20 hold-out split was used. Accuracy increased descriptively from 66.15% with all columns in the initial implementation to 73.54% with six selected features. However, the majority-class baseline that always predicts the very-low category achieved 80.16% accuracy. The selected-feature model also produced a macro F1-score of 0.24, balanced accuracy of 23.07%, and zero recall for the very-high category. Its 95% Wilson confidence interval for accuracy was 67.83%-78.56%, which does not establish superiority over the baseline. Cross-validation was not available, feature selection preceded data splitting, and an identifier column was included in the initial full-column implementation. Therefore, feature selection improved performance only relative to the initial Naïve Bayes model, but the resulting classifier is not yet valid, balanced, or suitable for clinical screening.