Using Nature - Inspired Metaheuristics to Train Predictive Machines

Nature-inspired metaheuristics for optimization have proven successful, due to their fine balance between exploration and exploitation of a search space. This balance can be further refined by hybridization. In this paper, we conduct experiments with some of the most promising nature-inspired metaheuristics, for assessing their performance when using them to replace backpropagation as a learning method for neural networks. The selected metaheuristics are: Cuckoo Search (CS), Gravitational Search Algorithm (GSA), Particle Swarm Optimization (PSO), the PSO-GSA hybridization, Many Optimizing Liaisons (MOL) and certain combinations of metaheuristics with local search methods. Both the neural network based classifiers and function approximators are evolved in this way. Classifiers have been evolved against a training dataset having bankruptcy prediction as a target, whereas function approximators have been evolved as NNARX models, where the target is to predict foreign exchange rates.