An automatic estimation of the ridge parameter for extreme learning machine.

Area of Science:

• Machine Learning
• Computational Intelligence
• Neural Networks

Background:

• Extreme Learning Machine (ELM) offers computational efficiency and fast training.
• ELM's real-time application often requires a ridge parameter (C) for ill-posed linear systems.
• Current methods for estimating C, like generalized cross-validation, are computationally expensive.

Purpose of the Study:

• To propose time-efficient methods for automatic ridge parameter estimation in ELM.
• To enhance generalization performance and computational speed for real-time ELM applications.
• To evaluate the effectiveness of L-curve and U-curve techniques for ridge parameter selection.

Main Methods:

• Developed novel methods utilizing L-curve and U-curve techniques for automatic ridge parameter estimation.
• Applied proposed methods to benchmark binary and multiclass classification datasets.
• Compared performance against existing methods using metrics like accuracy, precision, and F1-score.

Main Results:

• Proposed L-curve and U-curve methods effectively estimate the ridge parameter, even for large datasets.
• The new methods demonstrate superior performance in accuracy, precision, sensitivity, specificity, F1-score, and computational time.
• Statistical analysis using the Friedman ranking test confirms the proposed methods' effectiveness and superiority over existing techniques.

Conclusions:

• The L-curve and U-curve techniques provide a computationally efficient and accurate approach to automatic ridge parameter selection in ELM.
• These methods offer significant improvements in generalization performance and speed for real-time machine learning tasks.
• The proposed techniques represent a valuable advancement for the practical application of Extreme Learning Machines.