Walking to Authenticate: Identifying Robust Behavioral Biometrics from Step Count Data
Zhen Chen, Keqin Shi, Weiqiang Sun · 2023
Step count data, readily available from smartphones and wearable devices, offers valuable insights into a user’s physical activities and lifestyle patterns, positioning it as a viable contender for integration into biometric authentication systems. In this study, we introduce Step Count Print (SCP), a novel behavioral biometric derived from coarse-grained minute-level step count data, featuring daily step count distribution to capture an individual’s unique physical activity pattern. We conduct an ablation study with data from 100 users collected over a five-year period, utilizing mainstream machine learning algorithms. Our experimental results demonstrate SCP’s non-redundancy in user authentication scenarios, achieving an impressive average accuracy of $92.3 \%$. We showcase the universality and effectiveness of SCP across various classification algorithms through user accuracy histograms. Our proposed strategy holds promise for user-friendly and reliable biometric authentication, leveraging step count data to enhance security and usability in diverse applications.