Electronic Theses and Dissertation

Autism Spectrum Disorder (ASD) merupakan gangguan neurodevelopmental yang kompleks, ditandai dengan kesulitan dalam interaksi sosial, komunikasi, serta pola perilaku repetitif. Deteksi dini ASD sangat penting untuk meningkatkan kualitas hidup individu yang terdampak serta mengurangi beban bagi keluarga mereka. Penelitian ini mengusulkan sistem diagnosis berbantuan komputer untuk ASD dengan menerapkan arsitektur Vision Transformer (ViT-B/16) yang telah dilatih sebelumnya pada data sinyal EEG yang diperoleh dari King Abdul Aziz University. Dataset terdiri dari rekaman EEG dari 16 subjek (8 normal dan 8 ASD) yang telah melalui tahapan preprocessing, termasuk filtering menggunakan Discrete Wavelet Transform (DWT), segmentasi (windowing), serta konversi menjadi representasi heatmap, kemudian dibagi ke dalam subset latih, validasi, dan uji. Model ViT dilatih selama 100 epoch dengan batch size 16, menggunakan optimizer AdamW dan loss function CrossEntropy, serta mengevaluasi dua konfigurasi learning rate (0.0001 dan 0.00001). Bobot model terbaik dipilih berdasarkan nilai validation loss terendah. Hasil pengujian menunjukkan bahwa model yang dilatih dengan learning rate 0.00001 mencapai akurasi uji sebesar 99.53%, disertai dengan nilai precision, specificity, recall, dan f1-score yang sangat baik, sehingga menunjukkan kemampuan generalisasi yang kuat serta overfitting yang minimal. Penelitian selanjutnya disarankan untuk mengintegrasikan dataset yang bersumber secara lokal serta melakukan penyesuaian lebih lanjut terhadap arsitektur ViT melalui tuning hyperparameter yang lebih komprehensif, dengan tujuan mengembangkan aplikasi mobile guna mendukung diagnosis klinis ASD.

Kata Kunci: ASD, EEG, DWT, Visualisasi Heatmap, ViT.

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental disorder characterized by difficulties in social interaction, communication, and repetitive behavior patterns. Early detection of ASD is crucial for improving the quality of life of affected individuals and reducing the burden on their families. This study proposes a computer-aided diagnosis system for ASD by utilizing a pre-trained Vision Transformer (ViT-B/16) architecture on EEG signal data obtained from King Abdul Aziz University. The dataset comprises EEG recordings from 16 subjects (8 with ASD and 8 neurotypical), which underwent a preprocessing pipeline involving filtering using Discrete Wavelet Transform (DWT), signal segmentation through windowing, and conversion into heatmap representations. The data were then divided into training, validation, and test subsets. The ViT model was trained over 100 epochs with a batch size of 16, using the AdamW optimizer and CrossEntropy loss function, and evaluated with two learning rate configurations (0.0001 and 0.00001). The best model weights were selected based on the lowest validation loss. Testing results indicate that the model trained with a learning rate of 0.00001 achieved a test accuracy of 99.53%, along with excellent precision, specificity, recall, and F1-score values, demonstrating strong generalization performance and minimal overfitting. Future research is encouraged to incorporate locally sourced datasets and explore more comprehensive hyperparameter tuning of the ViT architecture, with the goal of developing a mobile application to support clinical ASD diagnosis. Keywords: ASD, EEG, DWT, Heatmap Visualization, ViT.