Identifikasi Autism Spectrum Disorder Pada Anak Berbasis Citra Ekspresi Wajah Menggunakan Arsitektur Resnet50 Dan Facial Action Unit

Kamilah, Silma (2026) Identifikasi Autism Spectrum Disorder Pada Anak Berbasis Citra Ekspresi Wajah Menggunakan Arsitektur Resnet50 Dan Facial Action Unit. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5002211064-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Autism Spectrum Disorder (ASD) merupakan gangguan perkembangan syaraf yang gejalanya mulai muncul pada usia 2-5 tahun, ditandai dengan hambatan dalam interaksi sosial dan ekspresi wajah yang atipikal. Diagnosis ASD saat ini bergantung pada observasi klinis dan wawancara subjektif sehingga berpotensi menyebabkan keterlambatan deteksi. Penelitian ini mengembangkan pendekatan komputasional skrining awal ASD berbasis analisis citra ekspresi wajah dengan mengintegrasikan fitur visual dari ResNet50 dan fitur perilaku wajah dari 18 Action Unit (AU) beserta entropinya. Dataset penelitian terdiri dari citra publik (TD-AFFECTNET dan ASD-FATMA) serta data video lapangan (TD-SDI dan ASD-SLB). Kerangka kerja penelitian ini diimplementasikan melalui dua pendekatan yaitu integrasi dengan Support Vector Machine (SVM) dan pembelajaran ResNet50 end-to-end. Evaluasi dilakukan melalui tiga skenario eksperimen dengan strategi pembagian data yang berbeda untuk menguji kinerja arsitektur model pada kondisi variasi sumber data serta pada kondisi data yang seragam. Hasil eksperimen menunjukkan bahwa integrasi AU dan ResNet50 secara end-to-end memberikan performa terbaik. Pada skenario pengujian yang melibatkan data pelatihan dari dataset publik dan data pengujian dari dataset lapangan, hasil mencapai akurasi sebesar 52,24%, yang dipengaruhi oleh perbedaan karakteristik antara data publik dan data lapangan. Sementara itu, pada skenario dengan data pelatihan dan pengujian dengan karakter yang sama, performa meningkat secara signifikan dengan akurasi sebesar 98,28%, F1-score sebesar 0,98, dan AUC sebesar 0,998. Penelitian ini membuktikan bahwa integrasi fitur visual dan perilaku mampu menghasilkan arsitektur skrining ASD yang objektif dan dapat diinterpretasikan, dengan potensi mendukung tenaga kesehatan dalam deteksi dini ASD.
=====================================================================================================================================
Autism Spectrum Disorder (ASD) is a neurodevelopmental condition whose symptoms typically begin to appear between the ages of 2 and 5 years, characterized by impairments in social interaction and atypical facial expressions. Current ASD diagnosis relies heavily on clinical observation and subjective interviews, which may lead to delays in detection. This study develops an early ASD screening system based on facial expression image analysis by integrating visual features extracted using ResNet50 and behavioral features represented by 18 Action Units (AUs) along with their entropy values. The dataset consists of public image data (TD-AFFECTNET and ASD-FATMA) as well as real-world video data collected in field settings (TD-SDI and ASD-SLB). The system is implemented using two approaches: feature integration with a Support Vector Machine (SVM) classifier and an end-to-end ResNet50 learning framework. Evaluation is conducted through three experimental scenarios with different data-splitting strategies to assess system performance under conditions of varying data sources as well as homogeneous data conditions. Experimental results show that the end-to-end integration of AU and ResNet50 achieves the best performance. In the scenario where training data are drawn from public datasets and testing data come from f ield-collected datasets, the system achieves an accuracy of 52.24%, which is influenced by differences in characteristics between public and field data. Meanwhile, in the scenario where both training and testing data originate from the same source, system performance improves significantly, achieving an accuracy of 98.28%, an F1-score of 0.98, and an AUC of 0.998. This study demonstrates that the integration of visual and behavioral features can produce an objective and interpretable ASD screening system, with the potential to support healthcare professionals in the early detection of ASD

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Autism Spectrum Disorder, Ekspresi Wajah, ResNet50, Facial Action Unit, Facial Expression
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.6 Computer programming. Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) R Medicine > RJ Pediatrics > RJ101 Child Health. Child health services
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Silma Kamilah
Date Deposited:	29 Jan 2026 03:23
Last Modified:	29 Jan 2026 03:23
URI:	http://repository.its.ac.id/id/eprint/130667

Actions (login required)

View Item