A hybrid dragonfly algorithm with extreme learning machine for prediction

In this work, a proposed hybrid dragonfly algorithm (DA) with extreme learning machine (ELM) system for prediction problem is presented. ELM model is considered a promising method for data regression and classification problems. It has fast training advantage, but it always requires a huge number of nodes in the hidden layer. The usage of a large number of nodes in the hidden layer increases the test/evaluation time of ELM. Also, there is no guarantee of optimality of weights and biases settings on the hidden layer. DA is a recently promising optimization algorithm that mimics the moving behavior of moths. DA is exploited here to select less number of nodes in the hidden layer to speed up the performance of the ELM. It also is used to choose the optimal hidden layer weights and biases. A set of assessment indicators is used to evaluate the proposed and compared methods over ten regression data sets from the UCI repository. Results prove the capability of the proposed DA-ELM model in searching for optimal feature combinations in feature space to enhance ELM generalization ability and prediction accuracy. The proposed model was compared against the set of commonly used optimizers and regression systems. These optimizers are namely, particle swarm optimization (PSO) and genetic algorithm (GA). The proposed DA-ELM model proved an advance overall compared methods in both accuracy and generalization ability.