Ensemble Merit Merge Feature Selection for Enhanced Multinomial Classification in Alzheimer's Dementia

Area of Science:

• Medical Informatics
• Machine Learning
• Neuroscience

Background:

• Diagnosing dementia is complex due to large feature sets from brain imaging and neuropsychological tests.
• Objective and reliable diagnostic techniques are crucial for effective dementia care.
• Conventional classification methods struggle with high-dimensional medical data.

Purpose of the Study:

• To develop an ensemble classifier with Merit Merge feature selection for efficient multivariate multiclass medical data classification.
• To enhance the accuracy and objectivity of disease diagnostics, specifically for dementia.
• To improve upon existing classification techniques for early and reliable disease detection.

Main Methods:

• Utilized an ensemble approach trained with features selected from multiple biomarkers.
• Employed Particle Swarm Optimization for feature subset retrieval.
• Experimented with classifiers including Naïve Bayes, Random Forest, Support Vector Machine, and C4.5.
• Integrated the C4.5 ensemble classifier with Particle Swarm Optimization search and Merit Merge technique (CPEMM).

Main Results:

• The proposed CPEMM feature selection method outperformed bagging feature selection across SVM, NB, and Random Forest classifiers.
• CPEMM identified an optimal feature subset for discriminating between normal individuals and patients with Mild Cognitive Impairment and Alzheimer's Dementia.
• Achieved a classification accuracy of 98.7% in distinguishing between healthy individuals and dementia patients.

Conclusions:

• The developed ensemble classifier with Merit Merge feature selection (CPEMM) significantly enhances classification efficiency and accuracy in medical data.
• This method offers a more objective and reliable approach to dementia diagnostics.
• The findings suggest a promising tool for early and accurate detection of Mild Cognitive Impairment and Alzheimer's Dementia.