Early Detection of Autism Spectrum Disorder Using Machine Learning on Speech Transcripts

Strategic Insights -

Researchers at the American University of Beirut have explored the feasibility of using machine learning on speech transcripts to detect Autism Spectrum Disorder (ASD) in children, aiming to improve developmental outcomes through early detection. The study's findings emphasize the potential of computational linguistics in advancing non-invasive, ethical approaches to ASD detection. By leveraging structured text-based inputs, the researchers minimized privacy risks associated with sensitive biometric data. Their machine learning models achieved accuracy above 86% on two datasets from the TalkBank repository, using a small, focused subset of features.

Key Takeaways:

  • The study aimed to develop a non-invasive, privacy-preserving approach to detecting ASD in children using machine learning on speech transcripts.
  • The researchers proposed an approach that leverages structured text-based inputs, minimizing the use of identifiable biometric data and associated privacy risks.
  • Experiments were conducted on two datasets from the TalkBank repository, focusing on linguistic features such as Mean Length of Utterance (MLU) and Mean Length of Turn Ratio (MLT Ratio).
  • The machine learning models achieved accuracy above 86% across both datasets, using a small, focused subset of features.
  • The study's findings highlight the potential of machine learning and computational linguistics in advancing ASD detection, providing a foundation for future applications in clinical and educational contexts.
  • The research was conducted by Rida Assaf, Zein Shehabeddine, and Vikram Ramesh at the American University of Beirut.

Statistics:

  • Accuracy of machine learning models: above 86% on both datasets.
  • Number of datasets used: 2 (from the TalkBank repository).
  • Linguistic features used: Mean Length of Utterance (MLU) and Mean Length of Turn Ratio (MLT Ratio).
  • Features used: small, focused subset of features, sufficient to maintain accuracy.
  • improvements in developmental outcomes: The study's findings suggest potential for improving developmental outcomes through early detection.

Sources:

  • Assaf, R., et al. (2025). Screening autism spectrum disorder in children using machine learning on speech transcripts. Scientific Reports, 15(1), 34134.
  • Nature Publishing Group. (n.d.). Scientific Reports. Retrieved from
  • American University of Beirut. (n.d.). Dept. of Computer Science. Retrieved from
  • NewsRx. (2025). Studies in the Area of Autism Spectrum Disorders Reported from American University of Beirut (Screening autism spectrum disorder in children using machine learning on speech transcripts). Pediatrics Week, 1159.