Effective Classification for Multi-modal Behavioral Authentication on Large-Scale Data

We propose an effective classification algorithm for machine learning to achieve higher performance for multi-modal behavioral authentication systems. Our algorithm uses a multiclass classification scheme that has a smaller number of classes than the number of users stored in the dataset. We also propose metrics, the self-mix-classified rate, other-single-classified rate, and equal-classified rate, for use with the proposed algorithm to determine an optimal number of classes for behavioral authentication. We conducted experiments using a large-scale dataset of activity histories that are stored when 100,000 users use commercial smartphone-applications to analyze performance measures such as false rejection rate, false acceptance rate, and equal error rate obtained with our proposed algorithm. The results indicate our algorithm achieved higher performance than that for previous ones.