Area based keystroke dynamics analysis for biometric authentication

In this thesis, a biometric authentication system for user identification through keystroke dynamics is discussed. A server-client system is developed to gather and model user data. Different from previous research on keystroke dynamics an area based feature is introduced to the classification model. Area based feature is developed and used as a discriminating feature via a virtual keyboard design. A software is developed for designing virtual keyboards and selecting areas. Experiments are conducted by designing different keyboards with different areas for classification of different users among several groups through machine learning techniques. User groups are formed from 10, 25, 50 and 100 persons. It is observed that accuracy is inversely correlated with group size. A higher accuracy is reported when area based feature is used. Top accurate results are obtained with area based features via Multilayer Perceptron classifier.