This title slide presents the project titled "Intelligent Bug Triage using Machine Learning for Agile Teams." It credits student Muhammad Zohaib Anwar, supervisor Dr. Mustafa Hameed, and the Department of Computer Science at [University Name].

Intelligent Bug Triage using Machine Learning for Agile Teams

Student: Muhammad Zohaib Anwar | Supervisor: Dr. Mustafa Hameed | Department of Computer Science, [University Name]

Source: Intelligent Bug Triage using Machine Learning for Agile Teams

This agenda slide, titled "Presentation Contents," outlines the structure of the presentation. It lists seven key sections: Introduction, Literature Review, Problem Statement, Research Questions, Research Objectives, Proposed Methodology Framework, and References.

Presentation Contents

1. Introduction
2. Literature Review
3. Problem Statement
4. Research Questions
5. Research Objectives
6. Proposed Methodology Framework
7. References

Source: DRC Research Proposal: Intelligent Bug Triage using Machine Learning for Agile Teams

This slide introduces bug triage, which assigns defects to developers in agile teams amid demands for swift handling in rapid iterations. It contrasts time-consuming manual processes with automation and ML for streamlined, intelligent prioritization and productivity gains.

Introduction

• Bug triage assigns defects to developers in agile teams
• Rapid agile iterations demand swift, efficient bug handling
• Manual triage is time-consuming and inconsistent
• Automation streamlines processes and enhances team productivity
• ML enables intelligent, scalable bug prioritization

Source: MS Research Synopsis

The literature review traces bug triage evolution from classical ML methods like SVM and Naive Bayes to deep learning and transformer models (e.g., BERT variants) for superior accuracy and feature representation. Key studies stress feature engineering improvements, alongside ongoing advancements in efficiency and scalability.

Literature Review

• Classical ML: SVM, Naive Bayes for initial bug triage.
• Evolution to deep learning for better feature representation.
• Transformers (BERT variants) boost accuracy significantly.
• Key studies emphasize feature engineering improvements.
• Ongoing advancements enhance triage efficiency and scalability.

The slide's "Problem Statement" lists challenges in bug triage, such as time-consuming manual processes and inconsistent assignments. It also notes knowledge gaps for new developers, scalability issues with growing projects, and workload imbalances across teams.

Problem Statement

• Time-consuming manual triage process
• Inconsistent bug assignments
• Knowledge gaps for new developers
• Scalability issues with growing projects
• Workload imbalance across teams

The slide "Research Questions" lists three key inquiries for an ML study. RQ1 asks if ML can achieve >70% Top-3 accuracy, RQ2 identifies the most important features, and RQ3 explores explainability for triage decisions.

Research Questions

• RQ1: Can ML achieve >70% Top-3 accuracy?
• RQ2: What features are most important?
• RQ3: How to provide explainability for triage decisions?

The research objectives aim to develop an ML model achieving over 70% Top-3 accuracy by comparing TF-IDF, Random Forest, XGBoost, and DistilBERT. They also include applying SHAP for model explainability and integrating a prototype for agile teams.

Research Objectives

• Develop ML model achieving >70% Top-3 accuracy.
• Compare TF-IDF, Random Forest, XGBoost, DistilBERT.
• Apply SHAP for model explainability.
• Integrate prototype for agile teams.

The proposed methodology framework outlines a six-phase workflow for bug triage, starting with dataset collection from Eclipse and Mozilla repositories, preprocessing via TF-IDF, and model development using Random Forest, XGBoost, and DistilBERT. It concludes with performance evaluation (Top-3 Accuracy, F1-score), explainability via SHAP values, and integration into a web prototype for agile teams.

Proposed Methodology Framework

{ "headers": [ "Phase", "Key Activities", "Techniques/Tools" ], "rows": [ [ "1. Dataset Collection", "Gather bug reports from public repositories", "Eclipse, Mozilla bugs" ], [ "2. Preprocessing", "Clean text data and feature extraction", "TF-IDF vectorization" ], [ "3. Model Development", "Train and tune classifiers", "Random Forest, XGBoost, DistilBERT" ], [ "4. Evaluation", "Assess model performance", "Top-3 Accuracy, F1-score" ], [ "5. Explainability", "Analyze feature importance and decisions", "SHAP values" ], [ "6. Prototype Integration", "Develop deployable tool for triage", "Web prototype for agile teams" ] ] }

Source: MS Research Synopsis

This slide serves as the section header for "Expected Outcomes" (Section 08). It outlines key deliverables: a high-accuracy triage model, interpretable SHAP insights, and a deployable prototype reducing triage time by 50%.

Expected Outcomes

08

Expected Outcomes

High-accuracy triage model, interpretable SHAP insights, deployable prototype reducing triage time by 50%.

Source: MS Research Synopsis

This slide, titled "References," lists six key academic citations in bullet points. They cover topics like software triage performance, defect prediction, BERT-based bug triage, software naturalness, Scikit-learn, and SHAP for model interpretation.

References

• Zhou, Y., Yang, S., Kaur, E., & Zhang, Y. (2012). Triage performance in rapid development.
• Lamkanfi, A., Verboonen, W., & Demeyer, S. (2015). Low effort defect prediction.
• Zhang, Y., et al. (2020). BERT-based models for intelligent bug triage.
• Hindle, A., et al. (2016). On the naturalness of software.
• Pedregosa, F., et al. (2011). Scikit-learn: Machine learning in Python.
• Lundberg, S. M., & Lee, S. I. (2017). A unified approach to interpreting model predictions (SHAP).

Source: MS Research Synopsis