fMRI-Based ADHD Diagnosis and Sex Prediction: A Machine Learning Approach

Here, we introduce a multi-output machine learning model for the joint classification of Attention-Deficit/Hyperactivity Disorder (ADHD) and biological sex based on functional MRI (fMRI) connectomes, socio-demographic characteristics, and behavioral measures of the WiDS Datathon 2025 dataset. The dataset contains more than 1,200 labeled samples, from 19,929 features to 100 principal components through PCA. We compare several classifiers such as Logistic Regression, Support Vector Machines (SVM), Random Forest, Gradient Boosting, XGBoost, LightGBM, and CatBoost. Experimental outcomes show that LightGBM has the best F1-score (72.53%), reflecting good classification performance with lower ROC-AUC scores compared to Logistic Regression and SVM, reflecting poorer probability calibration. The results indicate a trade-off between classification accuracy and probability ranking and have implications for clinical decision support. Future research will investigate probability calibration and ensemble methods to maximize both metrics, augmenting model applicability in risk stratification of ADHD. This enhancement is for the purpose of advancing individualized diagnostic tools as well as therapeutic approaches.