關鍵字：implicit authentication、使用者識別、遷移學習、領域適應、支持向量機、K-means、kernel map ;Nowadays, smartphone has been almost like a primary human needs. Almost everyone has minimum amount of one of smartphone. They tend to check on or use their smartphone at anytime and anywhere for many purposes. Therefore, security is also a need to secure people’s data or information inside their smartphone. Some implicit authentication method such as batch learning, active learning, and retraining have shown promising results to recognize the behavior of the user, including the one with unstable behavior. Behavior in our case is how the user flicks on the smartphones’s screen. The stable users tend to flick the same way all the time while the unstable ones tend to change their way to flick and it might be caused by certain reasons. An example of unstable behavior is when the unstable users might flick fast and long in certain time but might flick slow and short another time. The problem where some users tend to change their way of flicking and give data that are more scattered all over the sample space is called behavior change problem. In case of solving the problem with unstable user, we have found that retraining approaches which aim to retrain the classifier based on the new behavior of users showed a very good result to handle the unstable behavior. However, it takes a high computational cost and a long training time if dealing with relatively big data set. On the other hand, one of transfer learning’s approach parts which is the domain adaptation may share a similar ability with better efficiency in terms of training time and computational cost. This work presents non-instrusive behavioral biometric authentication on smartphones for behavior change handling based on domain adptation and empirical kernel map. The idea is to cluster the data sets so that recognizing user’s behaviors is easier and to form new feature representation for the data sets which is stable and less sensitive to behavior change. The experimental results show that the proposed approach has stable results and are better than initial baseline and Dyah’s work in terms of EER. It also shows that the proposed approach can handle the behavior change problem. Keywords: implicit authentication, user authentication, transfer learning, domain adaptation, support vector machine, k-means, kernel map
