近年來，智慧型手機被世界上大多數人廣泛使用， 因此考慮智慧型手機的資訊安全已經成為不可或缺的一環， 一些非侵入式方法包括批量學習、主動學習法、和再訓練在對應使用著上已經有著些有趣的結果， 在一些特別的情況下，像是使用者沒有固定的使用習慣，我們發現根據使用者的新行為來更新模型的再訓練方法可以有效的處理這些資料， 然而，當處理較大的資料集時，則是需要相當高的計算成本跟較長的訓練時間， 另一方面，轉移學習方法的一部份即所謂的領域適應性可以有著相似的效能，但有著更高的效率， 我們對無固定操作習慣之使用者提供了基於領域適應性之非侵入式手機使用者識別機制， 此方法是將目標數據從從目標域映射到原始域，因此我們可以將原始模型應用於目標數據， 實驗結果表示，跟基本方法相比，我們所提出的方法準確性更高，訓練和測試的時間也更快， 根據我們提出的方法能有效且滿足使用者行為的任何條件。;In recent years, the smartphone is widely used by most people in the world. Thus, the smartphone security has become a necessity since smartphones have increased in popularity. Several implicit authentication approaches include batch learning, active learning, and retraining have shown interesting results to map the behavior of the user, especially unstable user. In the particular case for resolving unstable user, we found that retraining approaches which aim to retrain the classifier based on the new behavior of users showed the good ability to handle the data. Moreover, it requires high computational cost and takes a long training time if dealing with larger dataset. On the other hand, one of the parts of the transfer learning approach which is so-called domain adaptation may share a similar ability with better efficiency. This work presents implicit behavioral authentication for unstable smartphone user based on domain adaptation. The idea of this approach is to map the target data from the target domain to the original domain thus we can apply the original model to the target data. The experimental result shows that the proposed method is better in term of EER compared to the basic approach and faster in term of training time than the retraining approach. However, the effectiveness yet satisfactory performances are letting this approach capable of handling any condition of user behavior data.
