A Hybrid Method Based on Extreme Learning Machine and Self Organizing Map for Pattern Classification

Area of Science:

• Artificial Intelligence
• Machine Learning
• Neural Networks

Background:

• Extreme learning machine (ELM) is a fast algorithm for single hidden layer feedforward neural networks.
• ELM performance is sensitive to the number of hidden neurons and random parameter initialization.

Purpose of the Study:

• To propose a novel hybrid learning method for optimizing ELM hidden neuron parameters.
• To enhance the accuracy and generalization capabilities of ELM.

Main Methods:

• A two-step hybrid learning process is introduced.
• Hidden layer parameters are adjusted using a self-organized learning algorithm.
• Output layer weights are determined via the Moore-Penrose inverse method.

Main Results:

• The proposed method was tested on nine classification datasets.
• It demonstrated superior performance compared to original ELM, Tikhonov regularization optimally pruned ELM, and backpropagation algorithms.
• The approach is both fast and effective.

Conclusions:

• The novel hybrid learning method offers significant improvements in ELM performance.
• It provides a robust solution for selecting optimal hidden neuron parameters.
• This method enhances both accuracy and generalization in classification tasks.