Comparison of Novel Random Forest Algorithm over Gaussian Naïve Bayes for Improving Accuracy in Software Bug Prediction System

This research work focuses on analyzing the performance of a proposed random forest (RF) method with that of Gaussian Naive Bayes in predicting software problems. The database utilized in this research was gathered from the Kaggle repository and contained information about software projects and their bug reports. A total of 20 samples were used in this study, with two groups formed and 10 samples randomly selected for each group. Group 1 was used for training and testing the Gaussian Naive Bayes algorithm, while Group 2 was used for training and testing the novel RF algorithm. Python and relevant libraries such as scikit-learn were utilized for implementation. The dataset was preprocessed to remove irrelevant features and perform feature engineering. The algorithms were tested on the training set, and their performance was assessed using accuracy metric on the testing data. A statistical power of 0.8 was set, and alpha and beta qualities of 0.05 and 0.2, respectively, were used with a confidence interval of 95%. The accuracy of the novel RF approach was 78.59% on average, whereas the accuracy of the Gaussian Naive Bayes approach was 75.57% on average. The p-value for the statistically distinct in accuracy among the both approaches was 0.001 ($\mathbf{p}\lt0.05$). The findings suggest that the novel RF approach could be a useful tool for software developers and project managers to predict and prevent potential bugs in their projects.