Improved barnacles mating optimizer algorithm for feature selection and support vector machine optimization.

Area of Science:

• Machine Learning
• Data Analysis
• Computational Intelligence

Background:

• Machine learning data analysis is crucial for complex data.
• Support Vector Machines (SVM) are powerful classification models facing challenges in feature selection and parameter tuning.
• Metaheuristic algorithms offer a solution for optimizing SVM parameters.

Purpose of the Study:

• To propose a novel classification model, IBMO-SVM, by hybridizing an improved barnacle mating optimizer (IBMO) with SVM.
• To enhance SVM performance through simultaneous optimization of feature subset selection and kernel parameters.
• To evaluate the effectiveness of IBMO-SVM on benchmark and real-world datasets, especially high-dimensional ones.

Main Methods:

• Development of an Improved Barnacle Mating Optimizer (IBMO) incorporating Gaussian mutation, logistic model, and refraction-learning strategies.
• Application of IBMO to optimize Support Vector Machine (SVM) feature selection and kernel parameters.
• Comparative analysis of IBMO-SVM against standard BMO-SVM and other state-of-the-art methods on 23 benchmark and 20 real-world datasets.

Main Results:

• The IBMO algorithm shows improved convergence accuracy and stability, balancing exploration and exploitation phases.
• IBMO-SVM significantly outperforms standard BMO-SVM and other compared methods on real-world datasets, particularly high-dimensional ones.
• The proposed IBMO-SVM demonstrates superior classification performance compared to four other classifiers.

Conclusions:

• IBMO-SVM is an effective hybrid model for improving SVM classification performance, especially in high-dimensional scenarios.
• The integrated optimization strategies in IBMO enhance the metaheuristic algorithm's effectiveness.
• This research contributes a robust and efficient machine learning model for complex data analysis challenges.