本文研究目的主要是以集成式技術中的Adaboosting作為分類評估的方法,此方法是用多個基底分類器做分類,再以不同權重方式投票,整合各項分類器之結果,得出一個較佳的模型。目的是比較單一分類器與結合多種分類器之優劣。本研究使用四種方法分別為簡單貝式、決策樹中的J48(c4.5)、支援向量機與倒傳遞類神經網路做為單一分類器,並使用Adaboosting將此四種單一分類器做為基底分類器互相結合,總共分為五大類型模型。使用UCI四個資料庫,葡萄牙銀行營銷、印地安人糖尿病、澳大利亞信用卡以及德國信用卡做為實驗的測試,並使用weka數據工具軟體進行測試。其研究結果顯示集成式演算法的分類結果較單一分類器來的準確,使用多個不同的分類器建立的集成架構相對於使用單一分類器,結果來的更準確更優良。
The aim of this research is to use the classifier ensemble method-Adaboosting to enhance the classification capability of the traditional classifier. The basic idea of Adaboosting is to include more than one base classifier in a classification procedure. The classification a results of the base classifiers are aggregated according to the different weights of the base classifiers. The classification performances of the Adaboosting based ensemble classifier will be compared with those of the single classifiers. This research employs four kids of base classifier techniques, including Naïve, Decision Trees, Support Vector Machines, and Back-Propagation Neural Network. Then the Adaboosting technique is used to construct the classifier ensemble using four base classifiers. Five kinds of models are created and compared. This research uses four databases of UCI as the test data. WEKA open software is used of as the platform to carry out the required experiments. The results shows that the Adaboosting based ensemble classifier can outperform the single classifier. Using more than one base classifier in a classifier ensemble can also improve the performance of the classifier ensemble.