EVALUATING THE EFFECTIVENESS OF DEEP LEARNING BASED ASSESSMENT IN VOCATIONAL EDUCATION: A META ANALYSIS

This study examines the effectiveness of a deep learning–based assessment system in vocational education, focusing on its implementation at SMK Negeri 8 Purworejo. To build a strong evidence base, the research first conducted a meta-analysis of earlier studies on automated deep learning assessment systems, selecting only those that used deep learning for performance-based evaluation, reported measurable accuracy or efficiency outcomes, and aligned with vocational or practical task contexts. The empirical phase tested a prototype system designed for productive Teknik Kendaraan Ringan (TKR) subjects, particularly for evaluating engine tune-up, component inspection, and basic troubleshooting practices. The system utilized video and image-based deep learning models to analyze student performance during hands-on activities. Data were collected from learning outcome documents, teacher interviews, student perception questionnaires, and observations of assessment duration. The results show that the deep learning–based system improves scoring precision, shortens evaluation time, and is positively received by both teachers and students. These practical gains support the meta-analytic findings and highlight the system’s potential to automate aspects of vocational assessment that were previously difficult to evaluate objectively. The study’s contribution lies in combining meta-analysis with real classroom implementation, offering a meaningful step toward AI-driven adaptive assessment in vocational education.