Efforts of Performance Optimization: The Experiment on Ten Accounting Datasets

Abstract

In the big data and digitalization era, fast-accurate decision-making has become a basic need, so data mining has a crucial role. The decision tree algorithm is quite commonly applied for classification functions, but performance level must always be evaluated for optimizing accuracy rate. Several optimization methods to accommodate these objectives include GA-bagging, PSO-bagging, forward selection, backward elimination, SMOTE, under-sampling, GA-Adaboost, and ABSMOTE-WIGFS. The results of the decision tree experiment on ten types of accounting-finance datasets used in this study obtained results with an average accuracy of 83.46%, an average precision of 65.64%, and an average AUC of 71.9%, while the majority of various optimizations are proven in improving the performance of decision tree algorithm where the application of ABSMOTE-WIGFS method is proven in providing the best rate with an average accuracy 87.71%, an average precision 87.09%, and an average AUC 84.87%, so it can be concluded that various optimization efforts are worth to be applied in case of accounting-finance themes for increasing the performance rate. Furthermore, the next research can prove these methods in other fields outside of accounting cases.