Deep Hybrid Intelligence: CNN-LSTM for Accurate Software Bug Prediction

Abstract

This research aimed to explore the potential of applying deep learning to software bug prediction. The study utilized various data preprocessing techniques that were essential in preparing the data for analysis, using a set of commonly available software bug reports and related metrics. In the data collection and preprocessing phase, the dataset was filtered to focus on critical software metrics, scaled for consistency, and additional techniques such as feature engineering and standardization were employed to enhance data variability. In order to analyze the effectiveness of the model in predicting software faults, the dataset was split so that it could be used for testing and training purposes. Several deep learning models, include CNN and LSTM architectures, were developed utilizing the preprocessed dataset in order to enhance the performance of the models. Subsequently, a hybrid ensemble technique was employed, combining the prediction outcomes of the best-performing individual models to form an ensemble model. Using test datasets, each model's performance was assessed using common assessment measures including precision, F1 score, accuracy, and recall. The ensemble models outperformed individual models in bug prediction, as demonstrated by higher accuracy and F1 scores. The final model achieved an accuracy of 96%, which was considered highly satisfactory for predicting software defects.