Exploring emotion prediction from biometric-based keystroke dynamics data using multiagent systems

The technologies of biometrics provide a variety of powerful tools to determine or confirm individual identity while, more recently, there has been considerable interest in using soft biometrics (personal information which is characteristic of, but not unique to, individuals) in the identification task. Increasingly, however, work has been developing to predict soft biometric information, such as predicting the age or gender of a subject, and this sort of process is clearly of particular interest in the context of criminal investigations. In this paper, we report some initial work to investigate the prediction of ”higher level” characteristics, specifically emotional state, of an individual from basic biometric data obtained from keystroke dynamics. We focus on the issue of specifying an underpinning computational platform based on a multiclassifier configuration and interacting agents to achieve better predictive performance than can be obtained using more traditional structures.