Optimized Laplacian Generalized Classifier Neural Network

A new variant of Generalized Classifier Neural Network (GCNN) is proposed. The traditional GCNN uses Gaussian RBF kernel where appropriate smoothing parameter is estimated using gradient descent based optimization. In this work, the Gaussian RBF kernel is replaced by Laplace kernel and optimal smoothing parameter is estimated using population-based optimization Sine Cosine Algorithm with a novel fitness function. This proposed method known as Sine Cosine Algorithm based Laplacian Generalized Classifier Neural Network (S-LGCNN). Different performance measures such as Accuracy, Specificity and F1-score are used to perform an analysis. Proposed method is tested on benchmark datasets and results are compared with other classifiers such as traditional GCNN, Logarithmic learning for GCNN, One pass learning for GCNN, probabilistic neural network and extreme learning machine. According to the test results, the proposed method can be considered an efficient way to improvise classification performance.