本計畫旨在研究多分類器系統的情境感知決策融合機制。給定一個包含類別標籤資料集，多分類器 系統的建置是訓練多個相異的成員分類器；給定一筆不含類別標籤的資料記錄當作輸入，每個成員 分類器的輸出是它對該筆記錄所隸屬的類別所做出的決策，而多分類器系統的運作是整合成員分類 器的輸出，或是融合成員分類器的個別決策以形成集體決策。多分類器系統的原始設計理念是透過 這樣的集體決策以達到更好的分類效能。其他學者已提出多個多分類器系統的決策融合機制，但這 些機制都只有考慮成員分類器的輸出而沒有考慮它們的輸入，亦即，這些機制都只有考慮成員分類 器的個別決策，而沒有考慮它們各自是在何種情境下做決策。因此，本計畫的研究問題可簡述如下： 在多分類器系統的決策融合過程中，如何考慮各個成員分類器的輸入及其採用的分類演算法的特性， 而不只是成員分類器的輸出，使得多分類器系統達到更好的分類效能？這是一個新的研究問題。就 最新學門規晝報告中提到的關鍵研究課題而言，我們可以將本計畫的研究問題直接連結到資訊融合 與集成學習。本計畫預期將對國内外的多分類器系統研究與應用做出實質貢獻，而研究成果將發表 於專業國際會議與期刊。 The purpose of this project is to study context-aware decision fusion mechanisms for multiple classifier systems. Given a data set that contains the class label, the construction of a multiple classifier system is to train several diverse member classifiers; given a data record that does not contain a class label as an input, the output of each member classifier is the decision it makes about to which class label the record belongs, and the operation of a multiple classifier system is to integrate outputs of members classifiers, or to fuse decisions of members classifiers in order to make a group decision. The original concept of multiple classifier systems is to achieve better classification performance through group decisions. Other scholars have proposed a number of decision fusion mechanisms for multiple classifier systems, but these mechanisms consider only outputs of member classifiers but not their inputs, that is, these mechanisms consider only individual decisions of member classifiers but not in which context they make decisions. Therefore, the research problem of this project can be briefly described as follows: In the decision fusion process of a multiple classifier system, how to take into account inputs of each member classifier and the characteristics of the classification algorithm adopted by it, but not just outputs of member classifiers, so that the multiple classifier system achieves better classification performance? This is a new research problem. Considering the key research topics mentioned in the latest Division Planning Report, we can connect the research problem of this project directly to information fusion and ensemble learning. Planning disciplines on the latest key research topics mentioned in the book, the research questions of this project may be directly related to information integration and integrated learning. This project is going to make practical contributions to research and applications of Multiple Classifier System, and the results will be published on professional international conferences and journals.