Repository of DUTH

Infant autism diagnosis via crying patterns analysis using machine learning (Master thesis)

Τσακαλίδου, Βικτώρια Νικολέτα/ Tsakalidou, Viktoria Nikoleta/ Προηγμένες Τεχνολογίες Πληροφορικής και Υπολογιστών

This master’s thesis explores the development and evaluation of a computational system for the early diagnosis of Autism Spectrum Disorder (ASD) in infants through the analysis of crying patterns. The primary objective of this research is to investigate the potential of using infant cry as a reliable biomarker, providing a non-invasive and accessible method for early intervention. The methodology is based on Machine Learning techniques. Initially, audio samples of cries from both infants diagnosed with autism and typically developing infants were collected and pre-processed. Subsequently, critical acoustic features were extracted, focusing on Melfrequency Cepstral Coefficients (MFCCs), fundamental frequency (F0), and voice quality indicators such as Jitter and Shimmer. For data classification, various algorithms were developed and compared, including Random Forests, Support Vector Machines (SVM), and XGBoost, with hyperparameter optimization performed to maximize accuracy. The main results of the study demonstrate that the proposed system achieves high success rates in identifying pathological crying patterns, confirming that sound analysis combined with artificial intelligence can serve as a powerful tool for clinicians in detecting autism at a very early stage.