Modelling of optimised neural network for classification and prediction of benchmark datasets
Pravin R. Kshirsagar, B. Nagaraj, Arpit Yadav
Abstract
Along with the reason of optimization of the Spread factor of a generalized regression neural network (Grnn) which is a variant of radial basis function network (Rbfn), the projected paper presents a novel algorithm which incorporates two of the soft-computing methods explicitly particle swarm optimization (Pso) as well as genetic algorithm (Ga). As in case of a normal Grnn for simulation of the network, instead of taking any random value for Spread factor, the Spread factor value has been optimized by using Pso then Ga as a result with the purpose of the network converges more rapidly with greater accuracy and minimum mean square error (MSE). Initial some random values of the Spread factor are calculated according to the positions in PSO algorithm and then the more unrealistic values of the spread are converted into realistic values using improved Ga. First on top of a few part of the UCI dataset Grnn is trained and this is used for testing on the remaining datasets. Projected algorithms classify and predict databases in the lesser amount of mean square error and with high accuracy also, the performance of the designed algorithm is tested by cross fold validation methods.