2025, Vol. 5, Issue 2, Part A

Emotion recognition in individuals with Autism Spectrum Disorder (ASD) has emerged as a critical area of research, given the challenges that individuals with ASD often face in recognizing and responding to emotional cues. This study explores the integration of sensing technologies and machine learning (ML) methods to advance emotion recognition in ASD. Emotion recognition in ASD is essential for improving social interactions and emotional intelligence, as individuals with ASD often exhibit deficits in processing facial expressions, body language, and vocal intonation. Despite significant advances in both sensing technologies and machine learning, a comprehensive solution for real-time emotion recognition that integrates these tools is still in its nascent stages. The aim of this study is to investigate how multimodal sensing technologies, combined with machine learning techniques, can enhance the accuracy and applicability of emotion recognition in individuals with ASD.

This research adopts a two-pronged approach to studying emotion recognition in ASD: (1) employing a range of sensing technologies, such as physiological sensors (heart rate variability, galvanic skin response), facial expression recognition via computer vision, and vocal tone analysis, and (2) leveraging machine learning methods, including supervised learning models (e.g., Support Vector Machines, Random Forests) and deep learning models (e.g., Convolutional Neural Networks, Recurrent Neural Networks) to analyze the data collected from these sensors. A dataset comprising individuals with ASD was created, capturing a wide array of emotional responses through these sensors in controlled environments. The performance of various machine learning algorithms was evaluated to determine their efficacy in recognizing emotions across different sensory modalities, and the integration of these modalities was explored to provide a holistic view of emotional states in individuals with ASD.

The results demonstrated a significant improvement in emotion recognition accuracy when using deep learning models compared to traditional machine learning algorithms. Facial expression recognition, when combined with physiological sensing, offered the highest accuracy, with models trained on this multimodal data reaching up to 90% accuracy in identifying emotions. Moreover, the study revealed that integrating multiple data sources resulted in more robust predictions, minimizing the influence of individual sensor variability. The findings also underscore the need for personalized approaches, as emotional responses can vary considerably between individuals with ASD, suggesting the necessity of adaptive systems that can tailor emotion recognition to specific behavioral patterns and emotional expressions.

In addition to the technical advancements, this research also highlights the practical applications of these findings. Real-time emotion recognition systems could be implemented in therapeutic settings, assisting therapists and caregivers in understanding emotional cues more effectively, thereby improving communication and intervention strategies. Furthermore, wearable devices that monitor emotions in real-time could be developed, offering continuous emotional support to individuals with ASD, particularly in environments like schools or public spaces where social interactions are frequent.

This study not only contributes to the technological advancements in emotion recognition for ASD but also opens new avenues for future research. Key areas of future exploration include refining algorithms to handle more diverse emotional expressions, addressing ethical concerns related to privacy and consent, and further integrating emotion recognition tools into everyday applications for individuals with ASD. The research highlights the need for more expansive datasets that encompass diverse age groups, genders, and severity levels of ASD to ensure the applicability and generalizability of emotion recognition systems.