No abstract

No abstract

Changes in software systems are inevitable. Identification of change-prone modules can help developers to focus efforts and resources on them. In this article, the authors conduct various intra-project and cross-project change predictions. The authors use distributional characteristics of dataset to generate rules which can be used for successful change prediction. The authors analyze the effectiveness of meta-heuristic decision trees in generating rules for successful cross-project change prediction. The employed meta-heuristic algorithms are hybrid decision tree genetic algorithms and oblique decision trees with evolutionary learning. The authors compare the performance of these meta-heuristic algorithms with C4.5 decision tree model. The authors observe that the accuracy of C4.5 decision tree is 73.33%, whereas the accuracy of the hybrid decision tree genetic algorithm and oblique decision tree are 75.00% and 75.56%, respectively. These values indicate that distributional characteristics are helpful in identifying suitable training set for cross-project change prediction.

Changes in software systems are inevitable. Identification of change-prone modules can help developers to focus efforts and resources on them. In this article, the authors conduct various intra-project and cross-project change predictions. The authors use distributional characteristics of dataset to generate rules which can be used for successful change prediction. The authors analyze the effectiveness of meta-heuristic decision trees in generating rules for successful cross-project change prediction. The employed meta-heuristic algorithms are hybrid decision tree genetic algorithms and oblique decision trees with evolutionary learning. The authors compare the performance of these meta-heuristic algorithms with C4.5 decision tree model. The authors observe that the accuracy of C4.5 decision tree is 73.33%, whereas the accuracy of the hybrid decision tree genetic algorithm and oblique decision tree are 75.00% and 75.56%, respectively. These values indicate that distributional characteristics are helpful in identifying suitable training set for cross-project change prediction.