Evaluating Civil Servant Selection through Machine Learning Analysis of National Insight, General Intelligence, and Personal Characteristics Test Scores

Fauzan Nur Adillah, Muhammad
Suakanto, Sinung
Ichsan Utama, Nur

Advance Sustainable Science, Engineering and Technology (ASSET)
Vol. 8 No. 2 (2026): February-April

This study analyzes the score distribution of 2,490 candidates in the 2024 Ministry of Finance Public sector recruitment, focusing on the CNI, GIT, and PCT sections using machine learning classification. Models used include Logistic Regression (accuracy 0.7897), Random Forest (0.9779), and XGBoost (0.9809), all trained with default parameters (n_estimators=100, max_depth=None) and evaluated using accuracy, precision, recall, and F1-score. While ensemble models outperformed Logistic Regression, the presence of false negatives—especially in the latter—reveals structural imbalances in test design. PCT scores dominate the total, while CNI and GIT show limited variation. These patterns suggest the need to revise PCT items with more complex ethical scenarios and enhance CNI and GIT content for better discrimination. This study contributes to improving test validity and fairness using empirical, data-driven methods. The findings support broader policy reforms toward more meritocratic and competency-aligned recruitment in Indonesia's civil service.