Fuzzy-based hunger games search algorithm for global optimization and feature selection using medical data.

Area of Science:

• Data Mining and Machine Learning
• Computational Intelligence
• Optimization Algorithms

Background:

• Feature selection (FS) is crucial for data preprocessing, aiming to reduce dimensionality, enhance model generalization, and improve classification accuracy.
• Traditional FS methods struggle with large search spaces, often failing to find optimal solutions.
• Hybrid techniques combining search strategies are common, but limitations persist.

Purpose of the Study:

• To propose a modified hunger games search algorithm (mHGS) for optimization and feature selection.
• To address drawbacks of the original hunger games search algorithm, including local search, premature convergence, and imbalance between exploration and exploitation phases.
• To evaluate the efficacy of mHGS on high-dimensional datasets for improving feature selection.

Main Methods:

• Development of the modified hunger games search algorithm (mHGS).
• Evaluation using the IEEE Congress on Evolutionary Computation 2020 (CEC'20) optimization test suite.
• Testing on ten medical and chemical datasets with dimensions up to 20,000 features or more.
• Comparison against established optimization algorithms like IMODE, GSA, GWO, HHO, WOA, SMA, and HGSO.

Main Results:

• The proposed mHGS demonstrated effective search results without increased computational cost.
• mHGS exhibited improved convergence speed compared to other methods.
• Experimental results showed enhanced Support Vector Machine (SVM) classification performance.

Conclusions:

• The modified hunger games search algorithm (mHGS) offers a robust solution for feature selection and optimization problems.
• mHGS effectively overcomes limitations of the original algorithm, providing better exploration-exploitation balance and faster convergence.
• The algorithm shows significant potential for improving classification performance on complex, high-dimensional datasets.