智慧型手機遊戲APP在Google的Google Play以及Apple的App Store兩平台之APP數量數量繁多,且遊戲類型APP競爭尤其激烈,因此APP的開發者莫不希望能夠儘量留住客戶,以創造最大的商業利益,因此若能預測可能流失用戶之情形,就能制定相關對策,降低顧客流失之速度。本研究以遊戲APP之使用者活動記錄為例,使用隱藏式馬可夫模型,建立依據時間序列所產生的使用者行為資料之使用者預測系統,藉由以該APP歷史資料所計算出之移轉機率矩陣與狀態值轉換機率,提供隱藏式馬可夫模型之訓練,並利用Viterbi演算法預測路徑之最終值為預測之狀態,建立使用者狀態之預測系統。實驗設計以用戶的持續活躍與可能流失之兩種情形為可能轉換之狀態,並依照遊戲APP使用者經常性會產生的活動為特徵值,進行模型的訓練與可能狀態之預測,提供APP的開發者可能流失用戶之名單,並採用十折交叉驗證與混亂矩陣之驗證方法,驗證系統之有效性,成功提供一個使用者可能行為狀態之預測系統。 Nowadays, you can find a lot of applications in Google’s Google Play and Apple’s App Store, and most of them are game applications. Therefore, all application developers are trying their best to keep users staying in application for developers’ interest. If we can predict the possible situation of losing users, we can develop countermeasures to slow down the pace.In this paper, we recorded users’ activities for example. Using Hidden Markov Model to build a prediction system by users’ behavior data generated from time series. By using the log of this application, to calculate the Transition probability Martix and Emission probability. Training Hidden Markov Model, and using the final route predicated by Viterbi algorithm as the predicated status to create a user prediction system.This experiment is base on two circumstance; users stay active and probably leave. According to game application users’ regular behaviors as the eigenvalues to train the modle and predicte the possible status, in order to provide a list of users that could possibly leave. Also, by using 10-Fold Cross Validation and Confusion Matrix authentication method to verify the system, can provide an effective prediction system to predict users’ behaveors.