Article

Neural Adaptive Learning Framework Based on Behavioral Sequences and Latent Cognitive States for Personalized Digital Instruction

---

Abstract

This study proposes a Neural Adaptive Learning Framework that integrates deep behavioral sequence modeling with latent cognitive state estimation to generate personalized instructional interventions in large-scale digital learning environments. Using a dataset comprising 9,000 learner behavioral sequences collected from an online learning platform, the framework employs a Bi-LSTM/Transformer encoder to model temporal dependencies across event logs and a latent-state inference module to estimate mastery, engagement, and cognitive load. Experimental results demonstrate that the model achieves an AUC of 0.89, outperforming a Random Forest baseline (0.76), a rule-based adaptive system (0.71), and a non-adaptive LMS (0.64). The adaptive policy also yields a 78% action accuracy, a substantial improvement over rule-based policies (55%) and non-adaptive sequencing (41%). In terms of educational impact, the framework leads to a 46% higher learning gain compared to the non-adaptive condition (0.41 vs. 0.28) and reduces time to mastery by 26% (9.1 vs. 12.4 sessions). Overall, the findings confirm that combining behavioral sequence encoding, latent cognitive inference, and neural policy optimization yields a robust, cognitively informed adaptive learning system. The proposed framework significantly enhances predictive accuracy, decision quality, and learning outcomes, offering a scalable and generalizable approach for next-generation personalized e-learning platforms.

Keywords: Neural Adaptive Learning, Behavioral Sequence Modeling, Cognitive State Estimation, Deep Learning, Intelligent Tutoring Systems

How to Cite:

Nugroho, K. A. & Turino, , (2025) “Neural Adaptive Learning Framework Based on Behavioral Sequences and Latent Cognitive States for Personalized Digital Instruction”, Adaptive Learning 1(4), 292-313. doi: https://doi.org/10.63913/al.v1i4.116