Improving Extreme Learning Machine Classification Using Adaptive Differential Evolution

Extreme Learning Machine (ELM) is the recently proposed learning approach which is based on Single-hidden Layer Feed-forward Network (SLFN). The main advantage of ELM is extremely fast learning speed and generalization ability but it suffers from randomly initialization of input weights and biases. To overcome this drawback, this paper proposes Adaptive Differential Evolution (ADE) method to find the best parameter setting of ELM. The proposed ADE utilizes the adaptation of mutation and crossover factors through the optimization process. The proposed ADE-ELM model is able to adaptively control the exploitation and exploration rates of DE over the search process, jump over the local optimal solution and obtain the global best solution (parameter setting of ELM) in the reduced running time. The results of the proposed method on several UCI datasets demonstrates the advantages of the proposed method.

Keywords:

ELM , Classification , ADE , adaptive mutation , adaptive crossover