Finding Initial Parameters of Neural Network for Data Clustering

K-means fast learning artificial neural network (K-FLANN) algorithm begins with the initialization oftwoparameters vigilance and tolerance which are the key to get optimal clustering outcome. The optimizationtask is to change these parameters so a desired mapping between inputs and outputs (clusters) of the K-FLANN is achieved. This study presents finding thebehavioral parameters of K-FLANN that yield goodclustering performance using an optimization methodknown as Differential Evolution. DE algorithm is asimple efficient meta-heuristic for global optimization over continuous spaces. The K-FLANN algorithmismodified to select winning neuron (centroid) for adata member in order to improve the matching rate frominput to output. The experiments were performed toevaluate the proposed work using machine learningartificial data sets for classification problems and synthetic data sets. The simulation results haverevealedthat optimization of K-FLANN has given quite promising results in terms of convergence rate and accuracywhen compared with other algorithms. Also the comparisons are made between K-FLANN and modified K-FLANN.