An early detection of learning disabilities using machine learning: A comparative study of models, datasets, and diagnostic strategies

Learning disabilities (LDs) affect a big portion of the school-age population, which often results in long-term academic and social difficulties. Early identification and timely intervention may lead to improved outcomes. The paper reviews the use of machine learning (ML) in the early detection of LDs, employing the PRISMA guidelines. The paper analyses peer-reviewed studies across different data models such as handwriting analysis, EEG, speech patterns, and behavior data. The review paper categorizes models by their architecture (e.g. SVM, CNN, hybrid models) and examines datasets, input features, model performances, and limitations. The paper concludes that hybrid and multimodal approaches (e.g., CNN+SVM, EFAM-XGB) offer superior accuracy. This paper identifies methodological gaps in the existing research and discusses the potential of the technology in future work, including the development of language–agnostic models, ethical concerns, and large-scale dataset standardization.