Data scarcity in adaptive learning: Challenges and opportunities with transfer learning and knowledge tracing

Abstract

Adaptive learning systems personalize instruction by tracking what each student knows and adjusting content to match their current knowledge state. Knowledge Tracing (KT) is the main technique used for this purpose. It has evolved from probabilistic models to deep learning architectures that use attention mechanisms to interpret a student's answer history and predict future performance. These models perform well only after training on large volumes of interaction data. A platform built for a new course, subject, student group, or national curriculum usually starts with little or no recorded history of student attempts. This is called the cold start problem. This review brings together recent work on knowledge tracing, transfer learning, and synthetic data generation to examine how these methods address data scarcity in adaptive learning. The review followed a systematic search and screening process across five academic databases, applied explicit inclusion and exclusion criteria, and used a structured quality appraisal completed independently by three reviewers. Seventy-nine records were identified; twenty-six were retained for the final synthesis (N = 26). The findings show that transfer learning techniques, particularly domain adaptation and cross-disciplinary knowledge transfer, can reduce the local data needed to reach acceptable prediction accuracy. However, none of the reviewed methods eliminate the cold start penalty; they only reduce it. The review also finds that most published work comes from a small number of research groups in the United States and China. Of the 26 included studies, 19 (73%) drew benchmark data or model architecture from these countries, leaving a gap for curricula, subjects, and languages elsewhere. Synthetic and simulated interaction data offer a complementary route: a new platform can begin with a working model before real student data accumulates. This research paper closes with guidance for researchers and developers building adaptive learning tools in data-scarce settings and directions for future study.