Towards a User-Friendly and Secure Hand Shaking Authentication for Smartphones

Abstract

Hand shaking authentication is a novel authentication method based on user behavior with mobile phone, however previous works commonly require users to repeat their actions dozens of times as dataset to train a machine learning model, causing unfriendly user experience. In this paper, we propose an efficient authentication method based on hand shaking, in which the user only needs to shake phones twice before use. The training set expands dynamically by adding samples when users successfully authenticate. A challenge is to achieve low FPR starting with very few training samples, which we referred as the cold start problem. In response, we present DTW-LSTM Online Stacking (DLOS), an ensemble authentication model which finely combines DTW with LSTM to create an accurate and robust classifier both on the cold start phase and the subsequent phase. The intuition is that DTW algorithm performs well enough with few samples while LSTM's performance improves significantly when the dataset grows. In DLOS, we introduce dataset_size φ as the meta-feature to online adjust weights of DTW and LSTM as the dataset grows. DTW plays a leading role in the cold start phase and LSTM dominates in the subsequent phase. Experimental results show that this method has a superior performance while saving the cost for usage, and it could effectively resist the shoulder surfing attack.