Abstract: A joint time-frequency analysis (JTFA) based method is proposed to improve the performance of the mouse dynamic trustworthy identity authentication system. Multi-resolution analysis of temporal signals based on wavelet packet transformation (WPT) is used to explore the joint time-frequency distribution characteristics of mouse behavior signals, and the recursive feature elimination method is applied to filter features extracted from users' long-term mouse behavior data. The random forest algorithm isemployed to establish the unique mouse behavior patterns for web users. which can be used to implement trustworthy identity authentication. In the case study, a single factor experiment is conducted, which sets the feature analysis method as the only factor. The mouse behavior data for thirty-one months of forty users is collected, among which the data of eighteen users are usedfor modeling and analysis. Both time series analysis and JTFA are adopted for feature extraction and the same algorithm is used to establish the model. The performance results show that the proposed JTFA based method increases the average value of area under receiver operating characteristic curve from 97.02% to 99.10% compared with the time series analysis based method.

Key words: joint time-frequency analysis, trustworthy identity authentication, mouse behavior, wavelet packet transformation, random forest