Summary: | 碩士 === 義守大學 === 資訊管理學系 === 103 === People nowadays have changed a lot in terms of life style and dietary habits, which tend to cause health problems and make the body sick. Not until this happens are people really willing to start paying attention to health issues more, trying to reduce the chance of falling ill. In this study, we apply decision tree theory of data mining in the prediction of breast cancer. After the analysis, we can verify which sorts of qualitative variables may mean cancer potential. This research has collected 699 data from Breast Cancer Wisconsin (Original) UCI; however, the exact number in use, except for the 16 data missed, is 683. The algorithms of J48, NB tree, Naive Bayes, Bayes Net, and Multi-Layer Perceptron implemented in the Weka software are also adopted to compare respectively and help get the accuracy of the results. Confusion matrix is used in categorization analysis. In addition, statistical analysis is applied in judging the importance of qualitative variables and finding the cause of breast cancer. Moreover, this study uses JFuzzLogic package, which programmed using Java language, and some theories related to regular decision analysis to develop a risk assessment system solely for detecting breast cancer. Through the reference of related knowledge and the use of C++, the study combine the different inference mechanisms used in this study and make them a basis for the user in both decision making and assessing. The results showed in this thesis can help people to make their own breast-cancer checks primarily and also can be a useful reference for a doctor when he/she diagnoses whether a patient with breast cancer or not.
|