A Multilayered Neural Network Based Computer Access Security System: Effects of Training Algorithms

摘要

In this paper, effects of training algorithms applicable to multilayered neural networks (NNs) are examined for a multilayered NN based computer access security system designed in order to differentiate an appropriate person from an intruder. Five training algorithms are taken into consideration for such system in terms of recognition accuracy. The algorithms studied are Backpropagation (BP), Quickprop (QP), Delta-Bar-Delta (DBD),Extended-Delta-Bar-Delta (EDBD), and Resilient Prop (RP). The designed system is trained using the data obtained from time intervals between successive characters while entering a password via keyboard. The performances of the algorithms are compared with each other in terms of classification accuracy.