An adaptive ensemble feature selection technique for model-agnostic diabetes prediction

Area of Science:

• Machine Learning
• Bioinformatics
• Data Science

Background:

• Ensemble learning enhances model performance by aggregating multiple models.
• Feature selection is crucial for identifying relevant predictors and reducing model complexity.
• Existing methods may not fully leverage the strengths of diverse feature selection techniques.

Purpose of the Study:

• To propose a novel Ensemble Feature Selection (EFS) method, named AdaptDiab, for diabetes prediction.
• To develop a model-agnostic approach that integrates various feature selection strategies.
• To enhance the accuracy and efficiency of predictive models through adaptive feature selection.

Main Methods:

• Developed AdaptDiab, an ensemble feature selection framework.
• Combined diverse feature selection techniques (filter and wrapper methods).
• Implemented an adaptive combiner function to dynamically select informative features based on ensemble member characteristics.

Main Results:

• Empirical studies demonstrated the effectiveness of AdaptDiab using various classification models.
• The proposed AdaptDiab method outperformed traditional feature selection methods.
• AdaptDiab provides a practical and improved framework for feature selection in ensemble learning.

Conclusions:

• AdaptDiab offers a significant advancement in ensemble feature selection for diabetes prediction.
• The model-agnostic nature of AdaptDiab allows for broad applicability across different classification models.
• This research contributes a robust and effective framework for optimizing feature selection in machine learning.