Ensemble-based Machine Learning and Deep Learning Approaches for Autism Prediction

Autism Spectrum Disorder (ASD) is a developmental disorder that affects a child’s cognitive capabilities and impacts social interaction and communication. Precise and timely diagnosis is very important, and while the initial techniques are based on observation, they do not yield consistent results due to their reliance on guesswork. In this paper, we develop an ensemble-based framework to boost the accuracy of autism prediction using a combination of machine learning and deep learning models. With this approach, Decision Trees, K-Nearest Neighbors, and Naive Bayes classifiers, as well as Multi-Layer Perceptron (MLP) and Recurrent Neural Network (RNN) models, are employed to exploit various characteristics of behavioral and diagnostic data. Each algorithm has its own degree of accuracy, and with ensemble learning, the system overcomes the disadvantages of uni-model predictions. This was validated experimentally, with the hybrid deep learning ensemble achieving 97% testing accuracy, which surpassed the results of individual models. The method promotes early detection by identifying behavioral signs associated with ASD patterns, allowing timely intervention that optimizes the treatment process. It also provides objective measurements, rapid processing, and high adaptability, which assist health professionals during clinical diagnostics, streamlining the decision-making process.