<![CDATA[Public vocational training effectiveness depends on alignment between training programs and job seekers’ professional profiles. In practice, training placement is often determined manually and subjectively, causing competency mismatches that reduce program effectiveness. This study develops an optimized computational framework to predict the suitability of vocational training programs for job seekers at the Department of Employment, Industry, and Trade of Batu Bara Regency. Using the Knowledge Discovery in Databases (KDD) framework, 1,434 historical records containing demographic data, education, work experience, occupational interests, and competency indicators were analyzed. To address the conditional independence limitation of the Naive Bayes classifier, three filter-based feature selection methods Information Gain, Mutual Information, and Chi-Square were implemented and compared. Results show that feature selection improved model performance, increasing accuracy from 91.26% to 93.01% across all methods. The consistent performance indicates that all methods identified the same dominant predictor, primarily professional interest, while removing redundant attributes. The proposed hybrid model demonstrates strong stability and generalization capability, providing a reliable decision support system for reducing employment mismatches and improving workforce development resource allocation.]]>
