Machine Learning Techniques for Detecting and Identifying Significant Autism Spectrum Disorder Speech Characteristics

Abstract

Autism spectrum disorder (ASD) is a group of neuro-developmental disorders that has a lifelong effect on social and communicative behaviors, and restricted and repetitive behaviors. Even though there is no cure, initial stage identification of ASD can help improve the patient’s life expectancy Conventional diagnosis of ASD is commonly performed through observation of behaviors and interview of a patient with medical experts by following standard procedures. In any case, these finding strategies are time-consuming. By integrating Machine Learning with neuroscience, a conclusion strategy might potentially be laid out to identify ASD subjects from typical development (TD) subjects. In this study, our main aim is to detect significant features that are associated with speech of ASD subjects by apply machine learning models on a cross-linguistic corpus (English and Danish. These mechanisms can help in earlier diagnosis The speech features from recordings are analyzed by various machine learning algorithms (like “Support Vector Machines (SVM), Random Forest Classifier (RFC), Logistic Regression (LR), XGBoost”) to build a classification model that classifies ASD and TD. In this process we achieved 97% accuracy with XGBoost in classifying ASD and TD. The model recognizes the important features in the order quasi-open-quotient (QoQ), jitter, pitch, duration, Dominant frequency (F2, F1, and F3), creak probability. Additionally, the findings support the use of voice and speech analysis as cutting-edge diagnostic tools for autism in young preverbal children.