OPTIMIZATION OF SOFTWARE DEFECT PREDICTION USING CNN AND ADABOOST: ANALYSIS AND EVALUATION

This study focuses on enhancing software defect prediction (SDP) by integrating Convolutional Neural Networks (CNN) with the AdaBoost algorithm. The PROMISE dataset was employed in this research, and data balancing was achieved using the SMOTE Tomek technique. With the help of AdaBoost, we were able to increase the prediction accuracy after building a complex CNN model to extract features from the da-taset. The AdaBoost model's hyperparameters were fine-tuned using GridSearch to find the best values for enhanced model performance. For the studies, we used StandardScaler to normalize the data after splitting it into training and testing groups with an 80:20 ratio. The ex-perimental results show that compared to the baseline method, SDP's accuracy is significantly improved when CNN, AdaBoost, and GridSearch hyperparameter tweaking are used together. Accuracy, pre-cision, recall, F1 score, MCC, and AUC were some of the measures used to assess the model's performance.