Opposition-Based Sine Cosine Algorithm (OSCA) for Training Feed-Forward Neural Networks

Abstract

Neural network is an effective machine learning technique for classification and regression. In recent studies many stochastic population based techniques are applied to train neural networks. In this paper, Opposition-Based Sine Cosine Algorithm (OSCA) is applied for feed-forward neural network (FNN) training. OSCA is a new population based metaheuristic, which is improved version of Sine Cosine Algorithm (SCA) and uses the opposition based learning (OBL) for better exploration. Performance is analysed and compared with Particle Swarm Optimization (PSO), Differential Evolution (DE), Genetic Algorithm (GA), Ant Colony Optimization (ACO) and Evolution Strategy (ES) for eight different datasets.