[Paper Review] An Evaluation of Machine Learning Approaches for Early Diagnosis of Autism Spectrum Disorder

Rownak Ara Rasul, Promy Saha+4 more|arXiv (Cornell University)|Sep 20, 2023

Autism Spectrum Disorder Research4 citations

TL;DR

This study evaluates eight supervised machine learning models and five unsupervised clustering algorithms for early autism spectrum disorder (ASD) diagnosis using children, adult, and combined datasets. Logistic Regression achieved 100% accuracy on the children dataset, while an optimized Artificial Neural Network reached 94.24% accuracy on the combined dataset; spectral clustering outperformed other clustering methods in NMI and ARI metrics, highlighting its robustness in label-free scenarios.

ABSTRACT

Autistic Spectrum Disorder (ASD) is a neurological disease characterized by difficulties with social interaction, communication, and repetitive activities. While its primary origin lies in genetics, early detection is crucial, and leveraging machine learning offers a promising avenue for a faster and more cost-effective diagnosis. This study employs diverse machine learning methods to identify crucial ASD traits, aiming to enhance and automate the diagnostic process. We study eight state-of-the-art classification models to determine their effectiveness in ASD detection. We evaluate the models using accuracy, precision, recall, specificity, F1-score, area under the curve (AUC), kappa, and log loss metrics to find the best classifier for these binary datasets. Among all the classification models, for the children dataset, the SVM and LR models achieve the highest accuracy of 100% and for the adult dataset, the LR model produces the highest accuracy of 97.14%. Our proposed ANN model provides the highest accuracy of 94.24% for the new combined dataset when hyperparameters are precisely tuned for each model. As almost all classification models achieve high accuracy which utilize true labels, we become interested in delving into five popular clustering algorithms to understand model behavior in scenarios without true labels. We calculate Normalized Mutual Information (NMI), Adjusted Rand Index (ARI), and Silhouette Coefficient (SC) metrics to select the best clustering models. Our evaluation finds that spectral clustering outperforms all other benchmarking clustering models in terms of NMI and ARI metrics while demonstrating comparability to the optimal SC achieved by k-means. The implemented code is available at GitHub.

Motivation & Objective

• To evaluate the performance of diverse supervised and unsupervised machine learning models in early ASD diagnosis.
• To identify the most effective classification and clustering algorithms for ASD detection using real-world datasets.
• To determine key clinical traits associated with ASD through feature importance analysis.
• To develop a GUI-enabled ML system for rapid, accessible ASD screening.
• To address the gap in comprehensive evaluation of ML approaches for ASD, especially in unsupervised settings without true labels.

Proposed method

• Applied eight supervised classification models—SVM, Logistic Regression, Random Forest, Gradient Boosting, Naive Bayes, K-Nearest Neighbors, Decision Tree, and Artificial Neural Network—on segmented and combined ASD datasets.
• Used hyperparameter tuning via grid search and Bayesian optimization to maximize model performance across multiple metrics.
• Evaluated classification models using accuracy, precision, recall, specificity, F1-score, AUC-ROC, kappa, and log loss for robustness assessment.
• Applied five unsupervised clustering algorithms—k-means, agglomerative clustering, Gaussian Mixture Models, spectral clustering, and BIRCH—on datasets without true labels.
• Measured clustering performance using Normalized Mutual Information (NMI), Adjusted Rand Index (ARI), and Silhouette Coefficient (SC).
• Developed a Streamlit-based GUI to enable real-time ASD risk prediction using input traits, enhancing clinical accessibility.

Experimental results

Research questions

• RQ1 Which supervised machine learning model achieves the highest diagnostic accuracy for ASD in children and adult populations?
• RQ2 How do different clustering algorithms perform when true labels are unavailable for ASD datasets?
• RQ3 What are the most predictive clinical features for ASD across different age groups?
• RQ4 Can a hybrid system combining ML models and a GUI enable faster, scalable, and cost-effective ASD screening?
• RQ5 How does model performance vary across segmented (children/adult) versus combined datasets?

Key findings

• For the children dataset, both SVM and Logistic Regression achieved 100% classification accuracy, indicating near-perfect detection performance.
• On the adult dataset, Logistic Regression achieved the highest accuracy of 97.14%, outperforming other models.
• The proposed Artificial Neural Network model achieved 94.24% accuracy on the combined dataset after hyperparameter tuning, demonstrating strong generalization.
• Spectral clustering outperformed all other clustering models in terms of NMI (0.539) and ARI (0.668) on the combined dataset, indicating superior cluster quality.
• The A9 score (aversion to physical contact) was identified as the most significant predictor in adult and combined datasets, while A4 (difficulty understanding others' emotions) was key in children.
• The developed GUI system enables real-time ASD risk prediction using input traits, offering a scalable tool for preliminary screening.