Neural-network's way to refine keystroke-information for high quality classification purposes
Anurag Tewari, Bipin Kumar Tripathi · 2021
A user’s typing manner defines identity of a user because typing rhythm of every user is always unique due to its link with behavior, style and mental attitude of user. On account of this, keystroke-feature creates a behavioral biometric recognition system. But timing based typing information can be deceptive also if information is shared by user half-heartedly. In this work neural-network-based classifier systematically investigates, presence of less significance in given dataset. Entries of few users are identified and eliminated from the data set, having relatively large misclassification errors. These entries are result of negative participation of some users during data collection process and during authentication process these entries may deteriorate performance-accuracy. This paper consists of different sets of password-entries based timings (of users) to be observed while classification procedure. Afterwards z–score normalization is applied to compare and analyze results of classification on original dataset and on reduced dataset.