Date of Award
Doctor of Philosophy (PhD)
Electrical Engineering and Computer Science
adversarial activity detection, behavioral biometrics, keystroke acoustics
Computer Sciences | Physical Sciences and Mathematics
Behavioral biometrics can be used in different security applications like authentication, identification, etc. One of the trending applications is predicting future activities of people and guessing whether they will engage in malicious activities in the future. In this research, we study the possibility of predicting future activities and propose novel methods for near-future activity prediction.
First, we study gait signals captured using smartphone accelerometer sensor and build a model to predict a future gait signal. Activity recognition using body movements captured from mobile phone sensors has been a major point of interest in recent research. Data that is being continuously read from mobile sensors can be used to recognize user activity. We propose a model for predicting human body movements based on the previous activity that has been read from sensors and continuously updating our prediction as new data becomes available. Our results show that our model can predict the future movement signal with a high accuracy that can contribute to several applications in the area.
Second, we study keystroke acoustics and build a model for predicting future activities of the users by recording their keystrokes audio. Using keystroke acoustics to predict typed text has significant advantages, such as being recorded covertly from a distance and requiring no physical access to the computer system. Recently, some studies have been done on keystroke acoustics, however, to the best of our knowledge none have used them to predict adversarial activities. On a dataset of two million keystrokes consisting of seven adversarial and one benign activity, we use a signal processing approach to extract keystrokes from the audio and a clustering method to recover the typed letters followed by a text recovery module to regenerate the typed words. Furthermore, we use a neural network model to classify the benign and adversarial activities and achieve significant results: (1) we extract individual keystroke sounds from the raw audio with 91% accuracy and recover words from audio recordings in a noisy environment with 71% average top-10 accuracy. (2) We classify adversarial activities with 93% to 98% average accuracy under different operating scenarios.
Third, we study the correlation between the personality traits of users with their keystroke and mouse dynamics. Even with the availability of multiple interfaces, such as voice, touch, etc., keyboard and mouse remain the primary interfaces to a computer. Any insights on the relation between keyboard and mouse dynamics with the personality type of the users can provide foundations for various applications, such as advertisement, social media, etc. We use a dataset of keystroke and mouse dynamics collected from 104 users together with their responses to two personality tests to analyze how their interaction with the computer relates to their personality. Our findings show that there are considerable trends and patterns in keystroke and mouse dynamics that are correlated with each personality type.
Fallahi, Amin, "Adversarial Activity Detection and Prediction Using Behavioral Biometrics" (2022). Dissertations - ALL. 1553.