Comparative Study of Machine Learning Based Defect Prediction Models for Python Software

Software Defect Prediction improves the software's stability and ensures the testing process is streamlined by pointing out issues in code. Software developers can manage and use their resources wisely, helping them maintain quality in what they produce by accurately forecasting defects. Several studies have been done on SDP, but most have concentrated on projects created in C, C++ or Java instead of Python. To fix this issue, we run different classical machine learning methods on the Python-based dataset, PyTraceBugs. Each code snippet in the dataset is assigned a label to indicate whether it has a bug and whether it is bug-free. It is a big issue that in this dataset, buggy instances never get close to matching non-buggy instances in number. To handle this, we select some of the buggy data randomly several times to make the model more accurate. A variety of evaluation metrics are used to determine how well the Random Forest, KNearest Neighbors, Decision Tree and Logistic Regression machine learning models work. In these cases, models can be evaluated using Accuracy, Precision, Recall, F1-score, AUC-ROC) and Matthews Correlation Coefficient (MCC). The analysis revealed that Random Forest is better at handling complex data patterns compared to other algorithms. These two approaches perform well and could be helpful for SDP tasks. When compared, Logistic Regression is far less effective, especially when identifying errors in code. Moreover, this study highlights that data preprocessing steps such as balancing classes and selecting features play a key role in improving defect prediction. The results build on existing Python-based research and provide guidance on putting ML models to use for spotting software errors.