Cross Project Defect Prediction Menggunakan Random Forest

Abstrak

This study develops a software defect prediction model using the Random Forest algorithm with a Many-to-One Cross-Project Defect Prediction approach. The model is tested using the AEEEM dataset as the training data source and the PROMISE dataset as the testing target. Both datasets consist of various software projects with important features such as code size, code churn, complexity, and other metrics that play a role in predicting software defects. The model’s performance is evaluated using metrics such as Accuracy, AUC, Recall, Precision, and F1-Score to measure its ability to detect defects across different datasets. The results show that the Random Forest model delivers excellent performance, with accuracy above 94% and AUC greater than 0.92 on most datasets. The model is also able to balance Recall and Precision effectively, resulting in more accurate and reliable predictions. Furthermore, this study applies ensemble techniques such as stacking and voting to combine predictions from multiple models, which significantly improve the stability and accuracy of the predictions. With this approach, the study demonstrates that the use of the Random Forest algorithm in CPDP can enhance the accuracy and efficiency of cross-project software defect prediction.