A Hybrid Feature Selection Approach for Early Detection of Alzheimer's Disease from Handwriting

The rapid increase in Alzheimer's disease patients due to global aging poses a significant public health concern, highlighting the urgent need for early diagnosis.This study proposes a hybrid feature selection technique to address the high-dimensional feature problem of the DARWIN (Diagnosis AlzheimeR WIth haNdwriting) handwriting dataset.The technique first extracts datasets with varying numbers of features using pre-selected feature selection techniques, experiments with six classification techniques, and then combines the two best-performing feature selection techniques into a hybrid approach.Experimental results identified Mutual Information and SHAP as the two optimal techniques, which were then combined to perform hybrid feature selection.Using the selected features, experiments demonstrated that the proposed technique achieved an accuracy of 93% with only 82 features and 92% with as few as 44 features.Future research will focus on the clinical relevance of the selected features and related tasks.