A novel chemical reaction optimization based higher order neural network (CRO-HONN) for nonlinear classification

(in lingua inglese)

In this paper, a chemical reaction optimized higher order Pi– Sigma neural network has been proposed. It is shown that the CRO can be able to optimize not only the structure of PSNN but also the weights of the network as it looks for the global minima of the objective function. CRO does not use any additional function or parameter such as fitness and low error rates than the other defined models. In case of statistical comparisons and performance analysis, the CROPSNN is statistically significant in all the tests. In ANOVA test, CRO-PSNN has mean value of 95.5129 (when a ¼ 0:05 in Post Hoc and Homogeneous test) and the upper bound value of 96.0713 is at 95% confidence interval.