Chain hybrid feature selection algorithm based on improved Grey Wolf Optimization algorithm.

Area of Science:

• Machine Learning
• Data Mining
• Bioinformatics

Background:

• Single feature selection methods often have limitations in effectiveness and performance.
• Hybrid approaches combining different techniques can overcome these limitations.
• There is a need for advanced algorithms to improve feature selection in complex datasets.

Purpose of the Study:

• To propose a novel hybrid feature selection algorithm, TMKMCRIGWO.
• To enhance classification accuracy and dimension reduction rates.
• To demonstrate the superiority of the proposed algorithm over existing methods.

Main Methods:

• The TMKMCRIGWO algorithm employs a two-stage filtering approach using Maximum Kendall Minimum Chi-Square (MKMC) and ReliefF.
• A wrapper algorithm, an improved Grey Wolf Optimization (IGWO) with random disturbance factors, is utilized for optimal subset selection.
• The tandem combination of filter and wrapper methods aims for robust feature selection.

Main Results:

• The TMKMCRIGWO algorithm achieved an average classification accuracy increase of at least 0.1% across 20 datasets compared to other algorithms.
• The average dimension reduction rate (DRR) reached 24.76%, with higher rates on low-dimensional datasets (41.04%) and lower rates on high-dimensional datasets (0.33%).
• The algorithm demonstrated improved model generalization ability and performance.

Conclusions:

• The proposed TMKMCRIGWO algorithm offers superior performance in feature selection compared to single methods.
• The hybrid strategy effectively addresses complex problems, leading to better classification accuracy and significant dimension reduction.
• TMKMCRIGWO shows promise for improving machine learning model performance and generalization.