Software Defect Prediction Using Atomic Rule Mining and Random Forest

Abstract

This research aims to improve software defect prediction in terms of accuracy and processing time. The new proposed algorithm is based on the Random Forest Algorithm that classifies and distributes the data based on tree module. It has value either 1 for defective module or 0 for the non-defective module. Random Forest Algorithm selects a feature from a subset of features which has been already classified. Random Forest Algorithm uses a number of trees for the prediction. For this research, datasets were tested with 10 and 15 sets of trees. Results showed an improvement in accuracy and processing time when the proposed system was used compared to the current solution for the software defect model generation and prediction. The proposed solution achieved an accuracy of 90.09% whereas processing time dropped by 54.14%. Processing time decreased from 19.78s to 9.07s during the prediction for over 100 records. Accuracy was improved from 89.97% to 90.09%. The proposed solution uses Atomic Rule Mining with Random Forest Algorithm for software defect prediction. It consists of classification and prediction process by using the Random Forest Algorithm during storing data that is carried out using Atomic Rule Mining.