Authenticating Smartphone Users Continuously Even if the Smartphone is in the User’s Pocket

Authenticating Smartphone Users Continuously Even if the Smartphone is in the User’s Pocket

연구 분야: Analysis

학회: International Conference on Advanced Computing, Machine Learning, Robotics and Internet Technologies

Users often underestimate the power and utility of smartphones, as they are unaware of the wide range of intelligent sensors that are now built-in into them. Smartphones are equipped with a wide array of sensors, including accelerometers, gyroscopes, magnetometers, barometers, pedometers, etc. These sensors are continuously active whenever the phone is turned on. Our study examines the way in which a smartphone can authenticate a user in a continuous and automatic manner, without requiring their awareness. To achieve this goal, we have created a moderate-sized human gait database that comprises data from 63 volunteers, taking into account their ages, genders, educational levels, professions, and backgrounds. In order to create this database, a web-based application was used to utilize the sensors on smartphones to capture human gaits. To determine which algorithm is most effective at verifying users based on their gait patterns, we analyzed the dataset here using ten popularly known anomaly detection algorithms. We found that one-class support vector machines (OCSVMs) are the most powerful for verifying users based on gait, with an average Equal Error Rate (EER) of 0.0043. It has been also identified that the EER has been reduced from 0.0043 to 0.0013 when soft biometric traits like gender, education level, and age group information are incorporated. We have proposed research directions to develop and implement this technology, and we believe it has practical applications.