AI-Powered Code Quality & Defect Analyzer

Analyze, visualize, and predict software defects using PROMISE research benchmarks, Git repository mining, static code metrics, and machine learning.

Features

• PROMISE Benchmark Explorer — Load standard defect-prediction datasets (Ant, Camel, jEdit, Lucene, Tomcat), explore CK metrics, and train classifiers.
• Git Repository Analyzer — Mine commit history for code churn and bug-fix labels, extract static metrics from Python/Java files, and predict defect hotspots.
• Single File Inspector — Paste or upload source code for real-time static analysis, defect probability scoring, and refactoring recommendations.

Tech Stack

Layer	Tools
Data collection	Python, GitPython
Static analysis	Radon (Python), regex-based parser (Java)
ML	Scikit-learn, XGBoost
Data / viz	Pandas, NumPy, Plotly, Matplotlib
UI	Streamlit

Project Structure

.
├── app.py                  # Streamlit dashboard
├── test_analyzer.py        # Integration tests
├── requirements.txt
├── data/                   # Cached PROMISE datasets (auto-downloaded)
└── src/
    ├── dataset_manager.py  # PROMISE dataset loading & preprocessing
    ├── git_miner.py        # Git history & churn mining
    ├── metrics_extractor.py# LOC, complexity, coupling metrics
    └── model_trainer.py    # LR / RF / XGBoost training & evaluation

Quick Start

1. Install dependencies

pip install -r requirements.txt

2. Run tests

python test_analyzer.py

3. Launch the dashboard

python -m streamlit run app.py

Open http://localhost:8501 in your browser.

Usage

Tab 1 — PROMISE Benchmark Explorer

1. Select a dataset (e.g. Apache Ant 1.7).
2. Explore defect distribution, metric scatter plots, and correlation heatmaps.
3. Choose a model (Logistic Regression, Random Forest, or XGBoost) and click Train Model Now.
4. Review ROC-AUC, confusion matrix, and feature importances.

Tab 2 — Git Repository Analyzer

1. Enter a local repo path or remote Git URL (presets include Flask and Requests).
2. Click Start Git & Code Analysis to mine commits and extract static metrics.
3. Optionally train a repo-specific defect predictor.
4. View defect hotspot predictions on a churn vs. complexity chart.

Tab 3 — Single File Inspector

1. Paste Python or Java code (or select a file mined in Tab 2).
2. Click Inspect Code Quality for LOC, complexity, coupling metrics.
3. If a model was trained in Tab 1 or Tab 2, see defect probability and refactoring advice.

Datasets

PROMISE datasets are downloaded automatically on first use and cached in data/:

Key	Project	Source
ant	Apache Ant 1.7	PROMISE-backup
camel	Apache Camel 1.6	PROMISE-backup
jedit	jEdit 4.3	PROMISE-backup
lucene	Apache Lucene 2.4	PROMISE-backup
tomcat	Apache Tomcat 6.0	DefectData

Metrics Extracted

Static (CK-style): LOC, WMC, DIT, NOC, CBO, cyclomatic complexity, method count

Git-derived: commit frequency, code churn, bug-fix count, author count

Requirements

• Python 3.10+
• Git (for repository mining features)

Author

Made by KAVYA RAJ