Improved Cuckoo Search in RBF Neural Network with Gaussian Distribution

Abstract

We propose an algorithm which improved our previous cuckoo search in RBF (Radial Basis Function) neural network by incorporating Gaussian distribution. Our algorithm can be used to forecast flood level in a river to assist in decision-making. Problems with training network in the classical method is that the best solution might unintentionally derive from local minima rather than the global one where we preferred. Hence, we train differently the parameters of neural network using a cuckoo search algorithm. The modified cuckoo search algorithm in RBF neural network uses Gaussian distribution in generating a cuckoo egg. We implement our algorithm using the actual water level from Little Wabash River as the input. The root mean squared errors of the previous results are improved by three new changes; increasing the number of input neurons, increasing the number of RBF nodes and changing the way the algorithm generated cuckoo eggs. Our results show that using cuckoo search via Gaussian distribution in the training phase of RBF neural network are preferable in both terms of time taken and prediction error.