Dispersed differential hunger games search for high dimensional gene data feature selection

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• High-dimensional genetic data presents challenges in clinical decision-making.
• Dimensionality reduction is crucial for efficient data processing and accurate classification.
• Existing gene selection methods may struggle with the balance between exploration and exploitation.

Purpose of the Study:

• To propose a novel wrapper gene selection algorithm, DDHGS, to address high-dimensional genetic data challenges.
• To enhance the search balance between exploration and exploitation in optimization.
• To improve classification accuracy and reduce computational costs in biological data analysis.

Main Methods:

• Developed the Dispersed Foraging Strategy and Differential Evolution combined HGS (DDHGS) algorithm.
• Introduced a binary derivative, bDDHGS, for feature selection.
• Evaluated DDHGS/bDDHGS against classic and advanced algorithms on IEEE CEC 2017 and 2014 benchmark suites.
• Tested bDDHGS on 14 UCI repository feature selection datasets.

Main Results:

• bDDHGS significantly outperformed existing methods, including bHGS, on 14 feature selection datasets.
• Demonstrated marked improvements in classification accuracy, reduced feature count, and better fitness scores.
• Showcased reduced execution time compared to other algorithms.
• bDDHGS exhibited superior performance on popular optimization functions and benchmark test suites.

Conclusions:

• bDDHGS is an effective feature selection tool in wrapper mode.
• The proposed DDHGS algorithm represents an optimal optimizer for high-dimensional biological data.
• This approach enhances the efficiency and accuracy of data-driven decision-making in medicine and biology.