“This thesis proposes an innovative mobile device user authentication based on modeling, learning and recognizing user device handholding patterns and finger gestures using machine learning algorithms of Hidden Markov Model, K-mean clustering and Least Mean Square. By developing an Android App on Google Nexus 5X, we collect two data sets from users to evaluate the system performance. The training process, which preprocesses the raw sensor data with the k-mean clustering algorithm, takes a short time to generate excellent performance for clustering.
An HMM algorithm is implemented in Java to learn and recognize a user’s gestures based on the preprocessed sensor data. With training with HMM, a common set of patterns are learned for each user. The trained HMM models are combined by Least Mean Square linear regression for the final authentication. The experimental results show that the system can achieve 89% accuracy. The accuracy can be improved with more sensor data taken for learning and training while a device is used for longer and longer”