A recognition approach using multilayer perceptron and keyboard dynamics patterns

Abstract

Multilayer perceptron (MLP) with one hidden layer is one of the most common forms of artificial neural networks ever utilized. A well-trained MLP with proper number of nodes in its hidden layer is demonstrated to have efficient and robust performance on patterns with high orders. In this paper in order to form an identification system, MLP is utilized as a classifier to distinguish keyboard dynamics patterns of several people. A variant number of neurons in the single hidden layer is investigated empirically to reach the optimum number. The optimum number of hidden layer neurons has been found to be 44 and relevant equal error rate (EER) equal to 0.95% has been reported. The false acceptance rate (FAR) and false reject rate (FRR) for this number of neuron has been empirically evaluated equal to 0.49% and 19.51% respectively.