The influence of emotions on touch behavioural features for biometric authentication

Abstract

Touchscreen devices have become increasingly popular recently, mostly due to the affordability and availability of smartphones and tablets. Smartphone security constitutes a necessary requirement due to the functions of smartphones that hold sensitive information and perform essential tasks. Numerous authentication techniques such as passwords, personal identification number codes, number locks, and graphical passwords are presently used to secure smartphones from unauthorised access. However, these techniques remain vulnerable to certain types of security breaches. To overcome the drawbacks of the current authentication techniques, behavioural biometric technology such as touch gesture authentication is being increasingly investigated. The touchscreen is a major source of data input, allowing users to make various movements such as scrolling, tapping, swiping, and so on. Touch gesture biometrics are identified as the process of computing and evaluating user touch gestures on touchscreen devices. When users interact with touchscreen devices, some forms of digital signatures are generated. These signatures may be used as an individual verifier because they are considered to be distinctive and unique for each user. Touch-based data collected from touchscreen sensors has been useful in various applications, such as emotion recognition, automotive vehicles, banking applications, signature verification, health care applications, gaming applications, and others. Recently, a number of studies have focused on using touch gestures as a form of biometric authentication for touchscreen mobile devices. However, these studies have faced several issues when developing touch gesture behavioural biometric approaches, mainly in improving the accuracy of the authentication system. Moreover, several behavioural factors such as emotions and their influences on touch gesture user authentication performance have remained unaddressed. In this research, the effect of emotions on user behaviour in influencing the performance of a touch gesture authentication approach was examined. To achieve this, a touch gesture behavioural biometric authentication approach was developed, and suitable experiment procedures were designed. Furthermore, a controlled experiment was conducted which allowed the collection of touch data in different emotional states (emotional and normal). An Android application was developed in order to collect the 572 touch gestures of 25 participants from touchscreen smartphones. The participants’ emotion states were induced using film clips’ emotion elicitation method and categorised based on the discrete emotion dimension (amusement, anger, sadness, tenderness, fear, and disgust). Eighteen touch features were extracted from the touch data and five machine learning classifiers were employed. Then, they were compared to evaluate the approach's accuracy. The results of the experiment indicate that the Random Forest technique achieved the best accuracy for the developed touch gesture authentication approach with 95.129% accuracy, 4.8% FRR, 0.22% FAR, and 2.5% EER. Furthermore, the influence of emotions was significant on the accuracy performance of the developed approach due to the accuracy value drop to 82.51%. Only 38.25% of the emotion datasets were correctly classified.