Hierarchical Harris hawks optimizer for feature selection.

Area of Science:

• Machine Learning
• Optimization Algorithms
• Data Science

Background:

• Wrapper-based feature selection methods rely on swarm intelligence algorithms, where performance is algorithm-dependent.
• Harris Hawks Optimization (HHO) offers fast convergence but struggles with complex, high-dimensional problems.
• Improving HHO is crucial for enhancing wrapper-based feature selection efficacy.

Purpose of the Study:

• To enhance the Harris Hawks Optimization (HHO) algorithm for improved feature selection performance.
• To develop an improved HHO algorithm, termed EHHO, capable of achieving high classification accuracy with fewer features and reduced computational time.
• To validate the effectiveness of the enhanced algorithm on diverse datasets.

Main Methods:

• Extensive experiments were conducted on 23 classical benchmark functions to evaluate EHHO against state-of-the-art metaheuristic algorithms.
• The EHHO algorithm was transformed into its binary variant, bEHHO, using a conversion function for feature selection tasks.
• The performance of bEHHO in feature extraction was rigorously verified on 30 UCI datasets.

Main Results:

• EHHO demonstrated superior convergence speed and minimum convergence values compared to other algorithms on benchmark functions.
• EHHO significantly mitigated the limitations of the original HHO algorithm when applied to complex functions.
• The binary EHHO (bEHHO) algorithm outperformed other optimization algorithms in feature selection tasks across 30 UCI datasets.

Conclusions:

• The enhanced bEHHO algorithm achieves excellent classification accuracy using fewer features compared to the original bHHO.
• bEHHO offers a significant improvement in running time over the original bHHO for feature selection.
• The study confirms EHHO's effectiveness in addressing complex optimization challenges in feature selection.