OPTIMIZING DEEP LEARNING MODELS FOR LIMITED DATA ENVIRONMENTS: A COMPARATIVE STUDY

Abstract

The use of deep learning models in education is expanding, particularly in supporting student data analysis, personalized learning, and AI-based evaluation tools. However, most of these models require large amounts of data to perform optimally, which often poses a challenge in educational environments with limited data. This study aims to explore and optimize deep learning models under limited data conditions through a comparative analysis of several approaches designed to improve model efficiency in such settings. It examines techniques like transfer learning, data augmentation, and semi-supervised learning, and evaluates model performance on educational data such as attendance records, exam scores, and student survey results from vocational high school students across West Jakarta. The findings reveal that transfer learning and data augmentation significantly enhance model accuracy without needing to directly increase data volume, while semi-supervised learning provides stable performance on highly limited datasets. These findings contribute to the development of more efficient deep learning models suited for educational environments with restricted data access, supporting educators and edtech developers in making informed decisions on the application of machine learning in educational institutions.