Answers are grounded in this paper and cite their sources.
Variance preservation does not necessarily establish meaningful achievement categories. This study compares six data transformations for principal component analysis (PCA)-based labeling using records from SMAN 1 Kertosono in 2024/2025. Academic data comprise 216 semester observations from 108 students in Grades XXII; non-academic data comprise 144 observations from 72 students in Grades XXI. Normalizer, StandardScaler, MinMaxScaler, MaxAbsScaler, RobustScaler, and QuantileTransformer were evaluated using student-grouped nested cross-validation. Transformations were selected by inner-validation macro F1 against teacher assessments. PCA fitting, sign orientation of the first principal component (PC1), and estimation of the one-third and two-thirds quantiles were confined to training data. These quantiles defined Low, Medium, and High categories for held-out students. MaxAbsScaler was selected in all five academic development sets, yielding 73.15% accuracy, macro F1 of 0.7233, and Cohen's kappa of 0.5982. Academic PC1 explained 81.0782.16% of transformed training variance, but labeling did not outperform the raw-score tertile baseline. Non-academic selections varied across folds, yielding 27.78% accuracy, macro F1 of 0.2377, and kappa of 0.0904. Non-academic PC1 explained 17.4827.86% of variance and primarily contrasted activity profiles, limiting its interpretation as an achievement scale. The contribution is an empirical evaluation separating variance retention, component interpretation, and teacher agreement. Results support a limited descriptive interpretation of academic PC1 but do not support the current non-academic categories as an overall achievement ranking. Findings are restricted to one school and require independent validation.
No takes yet. Share an insight, caveat, or question.
Waluyo et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: