A New Hybrid Approach For Classification Problem

Abstract

This work aims to introduce a new learning algorithm for classification issue. The suggested approach combine the self-organizing map (SOM) with the Extreme learning machine (ELM). In this algorithm, the load between input and hidden layers is made using the information retrieved from SOM on the training dataset. The weights of the output layer is adjusted applying an analytical method. Based on four classification benchmark, simulation results clarify that the new approach outperforms other learning algorithms and return sufficient performance in terms of learning speed and generalization. A comparative study with a number of other methods proves the efficiency of the proposed approach.