The Buggyness Predictor is a specialized software tool designed to assess the effectiveness of various machine learning classifiers in predicting the likelihood of a software class containing bugs. It provides a comprehensive suite of functionalities to import datasets, apply multiple classifiers, and analyze their performance based on key metrics. The two software analyzed are Bookkeeper and Avro from Apache.
-
Data import and preprocessing: the data is retrieved using the combination between Jira and Git. Missed information about the injected version and affected version are completed using proportion.
-
Classifier support: the classifiers analyzed are Random Forest, IBk and Naive Bayes. They are also combined with machine learning techniques such as feature selection, balancing and cost sensitive learning.
-
Performance and effort-aware metrics: Comprehensive evaluation metrics such as cost, precision, recall, F1-score, AUC and NPof30.
The results extracted from this evaluation can be found in the Documentation/ folder.
This tool is ideal for software developers, data scientists, and project managers looking to enhance their bug prediction models and improve software reliability.
Final project for the course Software Engineering 2, University of Rome Tor Vergata (Master's Degree in Computer Engineering), June 2024.