Heterogeneous Ensemble Combination Search Using Genetic Algorithm for Class Imbalanced Data Classification.

Area of Science:

• Machine Learning
• Data Science
• Computational Intelligence

Background:

• Classifying imbalanced datasets presents significant challenges in machine learning.
• Ensemble methods enhance classification but depend heavily on base classifier selection.

Purpose of the Study:

• To introduce a genetic algorithm-based optimization method (GA-EoC) for constructing effective heterogeneous ensembles.
• To address the challenge of imbalanced sample distributions in classification tasks.

Main Methods:

• Developed GA-EoC to search for optimal base classifier combinations for heterogeneous ensembles.
• Utilized 10-fold cross-validation for evaluating candidate ensembles and majority voting for combining decisions.
• Integrated random sub-sampling for class balancing and the (α, β) - k Feature Set method for feature selection when needed.

Main Results:

• GA-EoC demonstrated robust and superior performance compared to constituent base classifiers and other ensemble methods on benchmark, Alzheimer's disease, and PubFig datasets.
• The proposed method effectively handles class-imbalanced datasets.

Conclusions:

• Genetic algorithms provide a superior and reliable approach for heterogeneous ensemble construction.
• The GA-EoC method is expected to perform consistently across various classification scenarios.