An Improved Extreme Learning Machine Structure Based on Spherical Prototype Generator

Extreme learning machine (ELM) is a learning algorithm for the singlehidden layer feedforward neural network (SLFN) and has beensignificantly considered due to higher learning speed and moreefficient generalization power and excellent performance in regressionand classification issues, compared to traditional learning methods. Onthe other hand, determining the structure of ELM, which is equal todetermining the hidden layer parameters, plays a fundamental role inits performance. In the proposed method, a structure of ELM based onSPG method is presented to accurately regulate the parameters of thehidden layer and obtain a more compact structure of the hidden layer,so that the goal of effective classification of the data could beachieved. In general, the SPG algorithm improves the performance ofELM by selecting the most effective samples in the classification andhigh compression rate. The suggested method is applied to six datasetsof UCI and the results are compared to other methods (e.g., classicELM and FSVD-H-ELM) according to the evaluation parameter ofaccurate classification. According to the results of this study,application of the proposed method increases the accuracy of dataclassification, leading to improved performance of ELM structure.