Continuous biometric authentication typically relies on unstable and imperfect data input. Measuring the effectiveness of an authentication tool is in itself challenging due to the difficulty of modeling real word- settings in a test environment. The threat level, the average cost of an unauthorized access event and that of a false alarm all influence how the acceptance/rejection threshold should be calibrated to maximize effectiveness.
The variable nature of continuously observable biometrics such as cursor movement patterns, data noise and the availability of data itself, all impact the accuracy of continuous authentication systems. Machine learning models need to be trained for a relatively long time to balance the effect of inconsistent, unstable and inaccurate data. Data noise is created by both hardware and software related issues, like the mouse acceleration feature embedded at the operating system level, a tool designed to improve user experience.
There are also countless unpredictable factors that are inherent to cursor movement data, the basis of many continuous authentication solutions. Sitting or standing, using a mouse pad or not, mood changes, muscle strain, hormone levels or even a shot of espresso or a glass of wine all have an impact on what the authentication algorithm perceives through the cursor movement patterns.
Data science has answers to most of the highlighted challenges. As data processing algorithms improve, and with the possibility of combining different authentication methods, continuous biometric authentication will surely have growing importance in the foreseeable future.