An efficient alpha seeding method for optimized extreme learning machine-based feature selection algorithm.

Area of Science:

• Machine Learning
• Computational Science

Background:

• Embedded feature selection methods like Support Vector Machine Recursive Feature Elimination (SVM-RFE) are effective but computationally intensive.
• SVM-RFE faces challenges with model selection, leading to high computational burden and complexity.

Purpose of the Study:

• To propose an optimized extreme learning machine (OELM) model as a more efficient alternative to SVM for recursive feature elimination.
• To introduce an efficient active set solver for OELM training and an alpha seeding algorithm for successive quadratic programming (QP) problems.

Main Methods:

• Developed Optimized Extreme Learning Machine Recursive Feature Elimination (OELM-RFE).
• Implemented an efficient active set solver for OELM training.
• Introduced an alpha seeding algorithm to solve QP problems in OELM.

Main Results:

• OELM-RFE demonstrates higher prediction accuracy compared to SVM-RFE on benchmark datasets.
• OELM-RFE requires fewer model selection efforts due to having only one tuning parameter (penalty constant C).
• The alpha seeding method shows improved performance across a wider range of datasets.

Conclusions:

• OELM-RFE presents a computationally efficient and accurate feature selection method.
• The proposed alpha seeding algorithm enhances the efficiency of solving QP problems in OELM.
• OELM-RFE is a promising alternative to SVM-RFE for various machine learning applications.