The availability of educational data in novel ways and formats brings new opportunities to students with special education needs (SEN), whose behaviour and learning are highly sensitive to their body conditions and surrounding environments. Multimodal learning analytics (MMLA) captures learner and learning environment data in various modalities and analyses them to explain the underlying educational insights. In this work, we applied MMLA to predict SEN students’ behaviour change upon their participation in applied behaviour analysis (ABA) therapies, where ABA therapy is an intervention in special education that aims at treating behavioural problems and fostering positive behaviour changes. Here we show that by inputting multimodal educational data, our machine learning models and deep neural network can predict SEN students’ behaviour change with optimum performance of 98% accuracy and 97% precision. We also demonstrate how environmental, psychological, and motion sensor data can significantly improve the statistical performance of predictive models with only traditional educational data. Our work has been applied to the Integrated Intelligent Intervention Learning (3I Learning) System, enhancing intensive ABA therapies for over 500 SEN students in Hong Kong and Singapore since 2020.

Applied behavior analysis (ABA) has become a popular behavioral therapy in the special education needs (SEN) community. ABA is used to manage SEN students’ behaviors by solving problems in socially important settings, and puts emphasis on having precise measurements on physical and observable events. In this work, we present how Internet of Things (IoT) technologies can be applied to enhance ABA therapy in normal SEN classroom settings. We measured (1) learning performance data, (2) learners’ physiological data, and (3) learning environment sensors’ data. Upon preliminary analysis, we have found that learners’ physiological data is highly diverse, while learner performance seems to be related to learners’ electrodermal activity. Our preliminary findings suggest the possibility of enhancing ABA for SEN with IoT technologies.

We propose ABA-RAG, a retrieval-augmented generation (RAG) framework specifically tailored for applied behavior analysis (ABA) interventions, which integrates real-time emotional and behavioral data from Internet-of-Things (IoT) wearable devices. In ABA-RAG, we systematically retrieve semantically relevant prompts from a structured ABA task repository using embedding-based semantic search and dynamically construct context-aware prompts by incorporating IoT-derived predictions of learners’ emotional and behavioral states. To efficiently adapt large language model (LLM) to ABA contexts, we evaluate and compare several low-rank fine-tuning methods—including Low-Rank Adaptation (LoRA), Butterfly Orthogonal Fine-Tuning (BOFT), and LoRA with Gradient Approximation (LoRA-GA). Among these techniques, LoRA-GA demonstrates the best balance between computational efficiency and generation quality, making it particularly suitable for resource-constrained environments. Comprehensive experiments, validated through rigorous quantitative metrics and expert ABA practitioner evaluations, demonstrate that ABA-RAG significantly reduces manual workload while enhancing the precision, contextual relevance, and clinical utility of generated ABA tasks. Integrated into a practical web-based system, ABA-RAG provides ABA professionals with a scalable and real-time platform to generate individualized interventions.

Applied Behavior Analysis (ABA) therapy is a widely used intervention for students with special education needs (SEN), particularly those with Autism Spectrum Disorder (ASD) and co-occurring intellectual disabilities. However, despite its proven effectiveness, the integration of artificial intelligence (AI) and learning analytics (LA) in ABA therapy remains largely underexplored. This study examines the impact of an AI-driven learning analytics system on prediction performance, intervention effectiveness for SEN students, and support for therapists and teachers. The system collects and analyzes physiological, environmental, and behavioral data in real time to generate personalized intervention recommendations. A total of 33 students and 26 therapists/teachers from special schools and therapy centers in Hong Kong participated in an eight-week ABA intervention, followed by a post-evaluation session. The study assessed predictive accuracy, student learning outcomes, and educator perceptions using empirical data and qualitative feedback. Results indicate that the system achieved a predictive accuracy of 88.83% and a precision of 86.64% in forecasting learning outcomes, with statistically significant student performance improvement (medium effect size). Educators reported that the system's AI-driven recommendations enhanced their ability to develop individualized student profiles and intervention strategies. While the system did not replace traditional ABA methodologies, it improved decision-making by providing actionable insights through multimodal data integration. As of today, our system has been used by over 1,000 students with SEN in Hong Kong, Singapore, and Canada, demonstrating the real-world impact of AI-driven LA.

Biosignals collected from wearable devices are widely utilized in healthcare applications. Machine learning models used in these applications often rely on features extracted from biosignals due to their effectiveness, lower data dimensionality, and wide compatibility across various model architectures. However, existing feature extraction methods often lack task-specific contextual knowledge, struggle to identify optimal features in high-dimensional combinatorial feature space, and are prone to automated code generation and execution errors. In this paper, we propose DeepFeature, the first LLM-empowered, context-aware feature generation framework for wearable biosignals. DeepFeature introduces a multi-source feature generation mechanism that integrates the inherent ability of LLMs, expert knowledge and inter-feature interactions. It also employs an iterative feature refinement process that uses feature assessment-based feedback for feature re-selection. Additionally, DeepFeature utilizes a robust multi-layer filtering and verification approach for feature description-to-code translation to ensure that the feature extraction functions run without crashing. Experimental evaluation results show that DeepFeature achieves the highest average AUROC across eight tasks under both sample-level and subject-level settings, outperforming the best baselines by 4.60% and 4.61%, respectively. DeepFeature achieves the most pronounced gains on the PPG-BP tasks, while remaining competitive with the best-performing baselines on Epilepsy, WESAD, and our self-collected SEN dataset.

Emotion recognition from wearable physiology could enable timely, personalized support when behavioural cues are atypical or self-report is limited. This need is particularly acute for students with Autism Spectrum Disorder (ASD) during Applied Behavior Analysis (ABA) therapy, where therapists must often infer internal states indirectly and in real time. Although Transformer architectures excel at modelling complex temporal dependencies, most physiological emotion-recognition models are developed on neurotypical cohorts and are rarely evaluated under therapy-relevant validation protocols or integrated into operational ABA workflows. Here, we show that a supervised Transformer operating on continuous 1D peripheral physiological time series supports practical emotion recognition for ASD intervention and is associated with measurable improvements in therapy. Using therapist annotations strictly as training targets (not as input features), we benchmarked the model on an expert-curated dataset of 1,459,337 therapist-labelled samples from $n=12$ students with ASD collected during authentic sessions, achieving Leave-One-Session-Out (LOSecO) subject-specific accuracies of 84.28% (valence), 85.73% (arousal), and 80.26% (5-class emotion). In an independent user-study cohort ($N=33$, 1,722 sessions), integrating the system into ABA sessions was associated with statistically significant improvements in students' performance compared with standard practice. The system is further deployed in a production ABA platform serving over 2,000 students across 88 sites in Hong Kong, Singapore, and Canada, with 18 ms per-window inference latency and 99.97% uptime over three months. Our results demonstrate that Transformer-based physiological emotion recognition can be implemented reliably within the constraints of large-scale special education platforms, supporting practical translation from algorithmic performance to real-world intervention settings.