BiAffect: a System for Analyzing Neurocognitive Functioning Using Keystroke Dynamics and Machine Learning

Current techniques in diagnosing phases changes in patients affected by Bipolar disorder generally rely on daily hand-written questionnaires. These questionnaires have to be analyzed by doctors in order to find patterns or anomalies in the data that could indicate an imminent change in the mood of the patient. Moreover these results don’t represent objective psycho- physiological markers but, on the contrary, are highly biased by the perception of the patient himself. For this reason the current methods are not very accurate and cannot be considered a reliable support for the treatment of Bipolar patients. The proposed work aims at providing an efficient way to predict changes in the mood of patients and in order to allow doctors and relatives to promptly react and take appropriate countermeasures to them. However this work represents just an early presentation of the tech- niques that will be used in a future experimentation. Therefore its purpose is to present some general concepts that could be exploited in the future. In order to provide a confirmation of the validity of the considered concepts three main prototypes have been designed and imple- mented. The first prototype is a mobile application that manages the data collection process and which gives to the user the ability of choosing what to share with doctors and relatives. The second prototype is a keyboard application for smartphones that is able to log data com- ing from the patients, and the third is a visualization tool that gives the ability to doctors and experts to draw up preliminary analyses on the collected data. Promising results have already been obtained from the use of these prototypes. They show interesting discoveries about the keystrokes of the users. For instance it has been possible to determine with a discrete accuracy if a keystroke session have been performed with one or two hands. While this preliminary data is very promising, it doesn’t absolutely represent a conclusive point for this research. Still, these results are an encouraging signal for an actual validation of these techniques on real patients (which is forthcoming in the near future).