<![CDATA[The manual process of determining student eligibility for the Indonesia Pintar Program (PIP) often results in inefficiencies and inaccuracies. Schools are required to evaluate large volumes of socioeconomic data, and errors in judgment may lead to misallocation, where eligible students are excluded and ineligible students are included. Such inefficiencies highlight the need for objective, data-driven approaches. This study aims to evaluate the performance of the Naïve Bayes classification algorithm in classifying PIP eligibility, with a particular focus on attribute selection and its effect on classification accuracy. Historical student data from a primary school (SDN 1 Sindang Marga), which has rarely been examined in previous works and the analysis of attribute selection strategies, showing that fewer but more relevant attributes can yield better results. A dataset of 172 students was pre-processed and divided into training (80%) and testing (20%) subsets. Model evaluation was conducted using confusion matrices to calculate accuracy, precision, recall, and F1-score. The results demonstrate that using four attributes parental occupation, parental income, KPS ownership, and KIP ownership achieved the highest performance, with 85.3% accuracy, 92.0% precision, 88.5% recall, and a 90.2% F1-score. By contrast, using all seven attributes resulted in slightly lower accuracy (82.4%). These findings highlight that selective attribute use improves model efficiency and accuracy. Beyond methodological contributions, this research provides practical implications by demonstrating how machine learning can enhance fairness, transparency, and objectivity in educational aid distribution.]]>
