Statistical learning algorithms for dendritic neuron model artificial neural network based on sine cosine algorithm

Abstract

Training of dendritic neuron model artificial neural networks is generally achieved by using nonlinear least square methods. The distribution of random error terms is ignored in training algorithms although error terms are random variables. Maximum likelihood estimators can be obtained for dendritic neuron model artificial neural networks by using some indefinite symmetric probability distributions. In this study, statistical learning algorithms are proposed for dendritic neuron model artificial neural networks. Maximum likelihood estimators for dendritic neuron model artificial neural networks are obtained by using Normal, Cauchy, Logistic, Gumbel and Laplace distributions. The Sine cosine algorithm is used for maximization of the likelihood function under error terms that have Normal, Cauchy, Logistic, Gumbel and Laplace distributions. The proposed learning algorithms are applied to Istanbul Stock Exchange time series data sets. At the end of the analysis of application results, the performance of the proposed method is statistically better than well-known deep and shallow artificial neural networks.