Software Defect Prediction Using SMOTE and Artificial Neural Network
Wisnu Arya Dipa, Wikan Danar Sunindyo · 2021
defect prediction (SDP) is process of identifying software defect on the early testing stage of SDLC. SDP can saving time software tester on the development process. There are some issues on the way to develop SDP to be more effective, one of the issues is how to increase accuracy for predicting whereas most of dataset for SDP typically has imbalanced data for the defect class. In other words the dataset will naturally affecting appearance of prediction error on the classification model. This paper is to proposed Synthetic Minority Oversampling Technique (SMOTE) and artificial neural network to address the issues. The SMOTE is used to handling imbalance data and the artificial neural network is used to build predicting model. SMOTE and artificial neural network are applied to discover result of classification performance. The scenario is comparing imbalanced dataset that already processed with SMOTE and without SMOTE than classifying using artificial neural network and measured using value of precision, recall and f-measure. The experimental result show that the proposed SMOTE and Artificial Neural Network (ANN) and Association Rule Mining (ARM) method increase predicting performance for software defect prediction comparing with only Artificial Neural Network method. For parameter precision, recall, accuracy and F-measure improving increase are 2.2%, 18.4%, 22.4% and 8.8% At comparing ANN+ARM with ANN+ARM+SMOTE and beside that comparison about ANN+ARM with ANN+ARM+SMOTE also improve by 32.2%, 70.24%, 7% and 63.38%. The last comparison show different results from the others, the ANN+ARM+SMOTE get smaller performance value than combination ANN + SMOTE with gap score 17.2%, 7.8%, 1%, 15.6%. It happen because feature selection with Association Rule Mining didn’t help to improve predicting accuracy performance.