Applications of natural language processing and machine learning in autism spectrum disorder: a systematic review

Autism spectrum disorder (ASD) has been characterized as a spectrum, where patients struggle with communication, interpersonal interactions, sensory integration, and daily functioning changes, all of which are crucial for educational achievement. The integration of natural language processing (NLP) and machine learning (ML) technologies has shown promising results in various clinical prediction tasks, particularly in addressing early detection of diseases and illnesses. Furthermore, NLP techniques have been used to enhance the current assistive technology associated with autism. This systematic review aims to summarize the development and utilization of NLP in addressing autism issues by extracting unstructured data and classifying and analyzing data derived from clinical notes. A systematic literature review was conducted to assess the development of studies that used NLP to assist in the education and learning of children with autism. A total of 36 articles, including primary research article and systematic/scoping reviews published between 2016 and 2025 were included in this review. This work is divided into three main foci: (1) early detection of ASD by NLP technology, (2) classification of ASD via ML algorithms, and (3) implementation of ML in enhancing children with ASD learning experiences. Most of the review studies on ASD utilized electroencephalogram (EEG) signals to analyze the brain patterns of individuals with ASD. Furthermore, the most frequently used among ML models was support vector machine, which demonstrated excellent performance in classifying early detection of ASD in many of the reviewed studies. ML-based tools could help enhance the communication and social interactions of isolated individuals with ASD, but the reported results are diverse and mostly refer to small-scale or short-term interventions. For future work, researchers could explore various data sources because this will increase the accuracy of the ML model while reducing the risk of overfitting. The utilization of appropriate data sources can enhance the quality of future research by providing a highly comprehensive experimental validation analysis.