LAGAM: A Length-Adaptive Genetic Algorithm With Markov Blanket for High-Dimensional Feature Selection in...

Area of Science:

• Data Mining
• Machine Learning
• Computational Intelligence

Background:

• Feature selection (FS) is crucial in data mining for reducing dimensionality.
• Evolutionary computing (EC) shows promise for FS, but fixed-length encodings struggle with high-dimensional data.
• Existing EC-based FS methods face challenges like large search spaces and high computational costs.

Purpose of the Study:

• To propose a novel length-adaptive genetic algorithm with Markov blanket (LAGAM) for efficient feature selection.
• To address the limitations of fixed-length encodings in handling high-dimensional datasets.
• To enhance the search capability and efficiency of evolutionary algorithms in feature selection.

Main Methods:

• Developed a length-adaptive genetic algorithm (LAGAM) with variable-length individual encoding.
• Integrated an adaptive length changing operator to dynamically adjust individual lengths.
• Incorporated local search based on Markov Blanket (MB) to refine feature subsets.
• Rearranged features based on decreasing relevance to guide the search process.

Main Results:

• LAGAM demonstrated superior performance on 12 high-dimensional datasets compared to existing methods.
• Achieved higher classification accuracy while utilizing a significantly smaller number of features.
• The adaptive length encoding and MB integration proved effective in optimizing the search space.

Conclusions:

• LAGAM offers an effective and efficient solution for feature selection in high-dimensional data.
• The proposed adaptive encoding and MB integration enhance the performance of EC-based FS.
• LAGAM provides a promising direction for future research in evolutionary computation for data mining.