本研究以兩組資料不平衡問題的製造資料作為研究對象，資料來源分別為UCI

本研究以兩組資料不平衡問題的製造資料作為研究對象，資料來源分別為UCI資料庫的數據與某工廠的機台中實際收集，期望找出故障分類問題對於分類器績效的影響，並提供相關解決的方法，再進行各分類器之間的績效比較。本研究以減少多數法與SMOTE兩種採樣方法來處理資料不平衡的問題，搭配卡方特徵選取方法來找出有效提升分類器績效與訓練績效的特徵，並以決策樹、隨機森林、XGBoost與KNN四種機器學習演算法進行分類，經測試組資料的測試後，以正確率與AUC值作為績效評估的指標。The case group

結果 (英文) 1:

In this study, two groups of data manufacturing data imbalance as research subjects, sources UCI repository data were collected with the actual machine in a factory in Taichung, the impact is expected to locate the fault classification problem for classifier performance, and provide for the the solution, then the performance comparison between the classifiers.

In this study, the majority of law and reduce SMOTE two sampling methods to deal with the problem of data imbalance, with a chi-square feature selection methods to identify and effectively improve the performance of the classifier training performance characteristics, and to decision trees, random forest, XGBoost and KNN four kinds of machine learning algorithms to classify the test group after the test data, the correct rate and AUC values as indicators of performance evaluation.

結果 (英文) 2:

In this study, the manufacturing data of two sets of data imbalance problems are the research object, the data source is the data of the UCI database and the actual collection of the machine in a factory, hoping to find out the impact of fault classification problem on the classifier performance, and provide relevant solutions, and then carry out performance comparison between the classifiers.

This study reduces the problem of data imbalance by using most methods and SMOTE sampling methods, and uses the card feature selection method to find the characteristics that effectively improve the classifier performance and training performance, and classifies the four machine learning algorithms of the tree, random forest, XGBoost and KNN, after testing the test group data. , using the correct rate and AUC values as a pointer to performance evaluation.

結果 (英文) 3:

In this study, the manufacturing data of two sets of data imbalance problems are taken as the research object. The data sources are the data of UCI database and the actual collection of a factory's machine respectively. It is expected to find out the impact of fault classification problems on the performance of classifiers, and provide relevant solutions, and then compare the performance of each classifier.
In this study, two sampling methods, i.e. the majority method and the smote method, are used to deal with the problem of data imbalance, and the chi square feature selection method is used to find out the features that can effectively improve the performance of classifier and training performance. Four machine learning algorithms, decision tree, random forest, xgboost and KNN, are used to classify. After the test of test group data, the accuracy and AUC value are used as the indicators of performance evaluation Mark.