An Intelligent Emotion-Aware Learning Framework Using Temporal Facial Expression Recognition for Personalized Vocational Education
Facial Expression Recognition (FER) holds significant potential for real-time student emotion detection, directly influencing motivation, focus, and engagement in learning. In Indonesia’s dual-system vocational education where students alternate between classroom and industrial workshops an adaptive, context-aware FER system is essential. Most existing FER models rely on non-local datasets (e.g., FER2013, RAF-DB) and high-end hardware, limiting their applicability in resource-constrained Indonesian settings. This study proposes a lightweight, domain-adapted FER model using a hybrid architecture of ResNet18, Convolutional Block Attention Module (CBAM), and LSTM, optimized for standard laptops. The dataset combines FER2013, RAF-DB, and a local ASTRAtech dataset of 1,134 facial images collected via a Redmi Note 14 Pro+ 5G smartphone during authentic lecture and workshop sessions. Transfer learning and domain adaptation enable the model to recognize Indonesian students’ expressions under real-world challenges such as variable lighting and partial occlusion. The proposed system achieves 89.97% accuracy with 3.61 ms/frame latency (≈277 FPS) measured under controlled experimental conditions on standard laptop hardware, demonstrating near real-time performance suitable for adaptive learning applications. Integrated into an adaptive learning platform, it supports dynamic instructional adjustments based on detected emotions, enhancing pedagogical responsiveness. This research offers both a technical contribution a deployable lightweight FER model and a contextual contribution a validated approach to emotion-aware adaptive learning tailored for Indonesia’s vocational education ecosystem.
Authors
- Sasmito Budi Utomo
- AHMAD Abubakar
Institutions
- Federal University Dutsin-Ma (NG)
- Politeknik Astra (ID)
Publication Details
- Journal
- IJACI International Journal of Advanced Computing and Informatics
- Published
- 2026-11-20
- DOI
- https://doi.org/10.71129/ijaci.v3i1.pp42-52
- Primary Topic
- Emotion and Mood Recognition
- Type
- article
- Field-Weighted Citation Impact
- 0.00