Power of Ensemble Techniques for Brain Age Prediction Using Machine Learning Models

Abstract Aging has a profound impact on brain structure and function, resulting in cognitive decline and an increased susceptibility to neurodegenerative diseases. The brain age gap is defined as the difference between an individual's estimated brain age and their actual age, which is considered a potential marker of overall brain health and may indicate structural abnormalities. Considering this work, machine learning models are used to estimate the brain age based on brain imaging data from four prominent repositories namely IXI dataset, Calgary Campinas, Sparse Linear Method, and Sign Agnostic Learning with Derivatives respectively This work is done based on five regression models such as XGBoost (Extreme Gradient Boosting), Support Vector Regression (SVR), Gradient Boosting Regression (GBR), Random Forest Regression (RFR), and K-Nearest Neighbors (KNN) regression, and also considering ensemble models such as Stacking, Bagging, and Boosting. The dataset contains T1-weighted Magnetic resonance imaging (MRI) images from over 1800 patients, and approximately 143 features were extracted using the reliable tool FreeSurfer (6.0). In this study, a major analysis was performed using grid search and cross-validation to train the models and optimize the hyperparameters to prevent overfitting. The results exhibit that a stacked ensemble model of SVR and XGBoost outperformed the other models, with a mean absolute error (MAE) of 4.65 and R2 value of 0.92 on the training dataset while 6.62 and 0.85 on the test set respectively. The validation results indicated that regression models and ensemble techniques for brain age prediction provide a powerful combination of interpretability, accuracy, and robustness.