- 學位論文
貝氏網路於智慧家庭中行為預測之研究
A Bayesian Network Based Method for Activity Prediction in a Smart Home System
摘要
情境感知系統中，在辨識出使用者目前行為之後，若能預測居住人行為，就能提供更合適的服務給使用者。然而行為預測的研究通常將行為的預測結果視為唯一，但這樣的預測方式會導致覆蓋率不足的問題，為了解決上述的問題，提出預測的不同觀點和方法。在本論文中，以貝氏網路做為建立模型的方法，用來預測即將要發生的行為，並且將一般貝氏網路的預測結果做屬性的篩選，進而得到更多的預測結果。由於每位居住者的習慣有所不同，我們使用華盛頓大學CASAS的專案所提供的資料集中學習模型中相關參數，藉由此模型進行預測行為的動作，如此一來可以使得模型反映出真實世界的使用者習慣，進而更加貼近事實。最後，本論文所提供的方法可以用以提升行為預測模組的覆蓋率與預測準確率，並且與華盛頓大學CASAS的專案所提出的方法一起做比較，來驗證上述的提升。
並列摘要
In a context-aware system, if we can forecast what users will do, we can provide better services for users. Early research in activity prediction has indicted that the result of prediction is unique, but the question of coverage become apparent if only one result is considered. To solve this question, this thesis proposes a method with the view of multiple results. In this thesis, we use Bayesian Network to build a model that is used to predict which activity will happen, and then the predicted results will be applied to property filtering to get the final result. Due to the possibility that inhabitants may have different activity patterns, we have chosen the Bayesian Network to learn conditional probability of inhabitant activities from the data set of CASAS’s project. At last, the results are then compared to the result of CASAS’s project, and we find that the method in this thesis has improved coverage and accuracy in activity prediction.