Zulfan Zainal, Melinda Melinda, Yuwaldi Away, Marty Mawarpury
The diagnosis of autism spectrum disorder (ASD) using facial recognition technology faces challenges related to limited dataset variation and model classification effectiveness. The dataset is often less representative in capturing facial characteristics, lighting conditions, and social context, which impacts detection accuracy. This study aims to develop a new dataset with a broader variation in facial characteristics and refine the model classification based on Convolutional Neural Network (CNN) for optimal ASD detection. The developed dataset consists of 1,050 images with dimensions of 224 × 224 pixels, evenly distributed between the categories 'ASD' and 'Normal.' Evaluation using a nuisance matrix shows strong model performance, achieving 90% accuracy, 92% precision, 88% recall, and 90% F1 score. These findings suggest that a more representative dataset and image size optimization can improve the effectiveness of ASD detection. This study significantly supports a more reliable ASD diagnosis through machine learning technology. With promising results, this study paves the way for further application of facial recognition technology in clinical contexts and the potential development of automated detection systems that can be used in daily diagnostic practice. © 2025 IEEE.
School of Engineering, Universitas Syiah Kuala, Banda Aceh, Indonesia; Universitas Syiah Kuala, Faculty of Engineering, Department of Electrical and Computer Engineering, Banda Aceh, Indonesia; Universitas Syiah Kuala, Faculty of Medicine, Department of Psychology, Banda Aceh, Indonesia
Research at a Glance
Register to unlock
Topics & SDG Alignment
Register to unlock
Collaboration
Register to unlock
Author Profile (Selected)
Register to unlock
References Overview
Register to unlock
Journal & Source
Register to unlock
Metadata & Integrity
Register to unlock