Weighted Hybrid Random Forest Model for Significant Feature prediction in Alzheimer’s Disease Stages

• Published in: International journal of computational intelligence systems Vol. 18; no. 1; pp. 1 - 30
• Main Authors: Rohini, M., Surendran, D.
• Format: Journal Article
• Language: English
• Published: Dordrecht Springer Netherlands 10.03.2025; Springer Nature B.V; Springer
• Subjects: Algorithms; Alzheimer's disease; Artificial Intelligence; Brain; Classification; Cognition; Cognition & reasoning; Cognitive ability; Computational Intelligence; Control; Decision tree; Decision trees; Deep learning; Dementia; Density; Engineering; Impairment; Machine learning; Mathematical Logic and Foundations; MCI; Mechatronics; Prediction models; Random forest; Research Article; Robotics; Scaling factors; Weighted clinical feature; MCI; Random forest; Alzheimer’s disease; Decision tree; Weighted clinical feature
• ISSN: 1875-6883; 1875-6891; 1875-6883
• DOI: 10.1007/s44196-025-00780-0

In recent studies, several machine learning and deep learning prediction models have been proposed for the early detection and classification of various stages of Alzheimer’s Disease (AD). Many years before the actual onset of AD, there occur several structural changes in the brain. These structural brain features can be utilized in learning the disease progression from an early stage of the disease. The various stages of pathology cause mild cognitive impairment (MCI) from normal cognition and AD from normal cognition. This work intends to develop a weighted and hybrid random forest learning model that utilizes a relevant subset of predictors to diagnose the progression of the disease. The conversion from normal cognition to MCI is identified at an early stage of the onset of structural brain changes. The importance of proposed research works lies in more early identification of significant feature that increase disease progression and appropriate interventions greatly improve subjects’ recovery. The Alzheimer’s Disease Neuro Imaging Initiative (ADNI) cross-sectional MRI data were analyzed in this study that utilized brain curvature, grey matter density, white matter density, the volume of cortical and sub-cortical structures, shape of hippocampus, hippocampal subfield volume, Mini-mental state exam (MMSE), Clinical Dementia Rating (CDR), Estimated Total Intracranial Volume, Normalize Whole Brain Volume, and Atlas Scaling Factor for constructing randomized trees and thus predicting the features that cause the progression of disease stages from MCI to Alzheimer’s disease that causes dementia. Based on previous studies, there is a significant shortfall in understanding Alzheimer’s disease progression from pre-MCI stages and the classification of progressive and stable MCI groups. As a consequence of this challenge discussed, whether all the mild cognitively impaired people change to AD cohorts or remain in normal cognition and identification of the structural and functional features remains underexplored. Thus, the proposed Weighted Hybrid Random Forest algorithm (WHBM) utilized the 63 features that comprise the whole brain volume. The most significant and weighted features are derived which segregate 39% of subjects with cognitively progressive MCI and 51% of subjects with normal age-related cognitive decline. This implementation model proved to give robust AD conversion probability and identify significant features with 93% accuracy and 88% sensitivity that are sufficient for future clinical inferences. The optimized model thus resulted in the prediction of disease conversion probability from Mild Cognitive Impairment to AD because of significant structural features that are key-requisite for affected geriatric cohorts.