初步研究結果顯示，在決策樹中，未經資料平衡處理的分類器績效(正確率：0

初步研究結果顯示，在決策樹中，未經資料平衡處理的分類器績效(正確率：0.81%、AUC：0.57%)、若以減少多數法處理(正確率：0.66%、AUC：0.67%)、以SMOTE處理(正確率：0.82%、AUC：0.85%)。在隨機森林中，未經資料平衡處理的分類器績效(正確率：0.91%、AUC：0.5%)，經減少多數法處理(正確率：0.66%、AUC：0.63%)，經SMOTE處理(正確率：0.85%、AUC：0.87%)。由此可知，不同的採樣方法與分類器，皆會影響分類器的績效，因此，本研究期望找出績效表現最好的方法。

結果 (英文) 1:

Preliminary results show that, in the decision tree, without the classifier performance data processing balance (accuracy: 0.81%, AUC: 0.57% ), In terms of reducing most of the processing method (accuracy: 0.66%, AUC: 0.67% ) to SMOTE process (accuracy rate: 0.82%, AUC: 0.85% ).

In the random forest classifier performance data without balance processing (accuracy: 0.91%, AUC: 0.5% ), by reducing the majority of Treatment (accuracy: 0.66%, AUC: 0.63% ), by SMOTE treatment (right rate: 0.85%, AUC: 0.87% ).

It can be seen, different sampling methods and classification, will influence the performance of the classifier, therefore, expect this study to find out the best way to show performance.

結果 (英文) 2:

Preliminary results show that in the determination tree, the classifier performance (correct rate: 0.81%, AUC: 0.57%), if the majority method processing (correct rate: 0.66%, AUC: 0.67%), processing with SMOTE (correct rate: 0.82%, AUC: 0.85 %)。

In random forests, classifier performance (correctrate: 0.91%, AUC: 0.5%), reduced by most methods (correct rate: 0.66%, AUC: 0.63%), and SMOTE (correct rate: 0.85%, AUC: 0.87%).

It can be seen that different sampling methods and classifiers affect the performance of the classifier, so this study expects to find out the best way to perform.

結果 (英文) 3:

The preliminary research results show that in the decision tree, the performance of the classifier without data balance processing (accuracy: 0.81%, AUC: 0.57%), if the majority method processing is reduced (accuracy: 0.66%, AUC: 0.67%), and smote processing (accuracy: 0.82%, AUC: 0.85%).
In the random forest, the performance of the classifier without data balance processing (accuracy: 0.91%, AUC: 0.5%), by reducing the majority of methods (accuracy: 0.66%, AUC: 0.63%), by smote processing (accuracy: 0.85%, AUC: 0.87%).
It can be seen that different sampling methods and classifiers will affect the performance of classifiers. Therefore, this study is expected to find the best performance method.