Figure 1.

Overview of BiAffect data collection and feature extraction process. (A) Keypress-level typing metadata are collected via the BiAffect keyboard and stored by Sage Bionetworks. (B) Interkey delays (IKDs) for keypress transitions from character to character are aggregated at a session level to obtain the 25th, 50th (median), and 95th percentile IKDs and the median absolute deviance (MAD) IKD. Typing accuracy in a session is defined as the autocorrect rate (ie, ratio of autocorrect instances to the total number of characters). Last, the session duration (seconds) in each session is obtained by aggregating all delays in a session. (C) An example for the hourly typing activity over multiple days from 2 active users is shown in the top right as an illustration of the potential patterns captured via continuous, unobtrusive collection. The blue dashed line highlights the different levels of activity at night, with user B exhibiting a more irregular activity pattern than user A. Size of the marker is proportional to the number of characters typed per hour.