Deep ensemble learning for Alzheimer's disease classification.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Computational Neuroscience

Background:

• Ensemble learning enhances predictive performance by combining multiple algorithms.
• Deep learning is increasingly used for complex tasks, but its application in Alzheimer's disease classification ensembles is limited.
• Integrating multisource data is crucial for accurate Alzheimer's disease diagnosis.

Purpose of the Study:

• To present a novel deep ensemble learning framework for Alzheimer's disease classification.
• To integrate multisource data effectively using deep learning algorithms.
• To improve the accuracy and reliability of Alzheimer's disease diagnostic services.

Main Methods:

• A deep ensemble learning framework incorporating sparse autoencoders for feature learning and a deep belief network for classifier ranking.
• Utilizing a neural network as a meta-classifier.
• Employing over-sampling and threshold-moving techniques to address cost-sensitive classification problems.

Main Results:

• The proposed deep ensemble framework achieved 4% higher classification accuracy compared to six established ensemble methods, including standard stacking.
• Demonstrated improved performance on a clinical dataset from the National Alzheimer's Coordinating Center.
• The framework effectively reduces attribute correlation and diversifies base classifiers.

Conclusions:

• The developed deep ensemble learning framework offers a significant advancement in Alzheimer's disease classification.
• This approach provides a new avenue for enhancing Alzheimer's disease primary care through machine learning.
• The findings suggest potential for improved diagnostic services by aggregating insights from multiple computational 'experts'.