Code Quality Assessment Model

An AI/ML project that analyzes software metrics to predict code quality.
This project explores how static code features and software metrics can be used to build a machine learning model that estimates the quality of source code.

---

Project Overview

Software quality is an important part of software engineering, especially when trying to detect maintainability issues, code smells, and poor design patterns.
This project aims to use machine learning to classify or estimate code quality based on measurable code metrics.

The workflow includes:

• Loading and exploring the dataset.
• Cleaning and preprocessing the data.
• Engineering relevant features.
• Training machine learning models.
• Evaluating performance with standard metrics.
• Saving the final model for reuse.

---

Objectives

• Build an AI model for code quality assessment.
• Analyze the relationship between software metrics and code quality.
• Compare different machine learning algorithms.
• Create a reproducible pipeline for experimentation and evaluation.
• Produce a GitHub-ready project that demonstrates AI/ML and software engineering skills.

---

Dataset

This project uses the Software Code Quality and Source Code Metrics Dataset from Mendeley available in this link : https://data.mendeley.com/datasets/77p6rzb73n/5 The dataset contains synthesized code smells, issues, quality, and source code metrics information of 60 versions under 10 different repositories.

Dataset Contents

The version 5 is used. The dataset includes these following information:

• Source code metrics
• Code smells
• Static analysis issues
• Quality attributes

Analysis Levels

The data is available at three levels:

• Package level
• Class level
• Method level

Key Files

• repositories.csv – Repository information (name, URL, commits, stars, forks)
• versions.csv – Version information and repository growth metrics
• attribute-details.csv – Description of quality attributes and metrics
• codesmells.csv – Code smell rules and descriptions

Dataset Scale

• 9.4 million+ lines of code analyzed
• 173,000+ classes analyzed
• 10 repositories
• 60 software versions

The full dataset is saved in the data/raw/ folder.

---

Problem Type

This project can be formulated as:

• Classification, if the target is a quality class.
• Regression, if the target is a numeric quality score.

In the first version of the project, a classification approach has been chosen for simplicity and clearer interpretation.

---

Project Structure

code-quality-assessment-model/
├── README.md
├── requirements.txt
├── .gitignore
├── LICENSE
├── data/
│   ├── raw/
│   ├── processed/
│   └── sample/
├── notebooks/
│   ├── 01_data_exploration.ipynb
│   ├── 02_preprocessing.ipynb
│   ├── 03_model_training.ipynb
│   └── 04_model_evaluation.ipynb
├── src/
│   ├── __init__.py
│   ├── config.py
│   ├── data_loader.py
│   ├── preprocessing.py
│   ├── feature_engineering.py
│   ├── train.py
│   ├── evaluate.py
│   └── predict.py
├── models/
│   ├── trained_model.pkl
│   └── label_encoder.pkl
├── reports/
│   ├── figures/
│   └── metrics/
├── tests/
│   ├── test_preprocessing.py
│   ├── test_model.py
│   └── test_prediction.py
└── docs/
    ├── project_overview.md
    └── dataset_description.md

---

Methodology

1. Data Exploration

The dataset is loaded and inspected to understand:

• Column names.
• Missing values.
• Data types.
• Distributions of features.
• Target label availability.

2. Data Cleaning

The preprocessing step includes:

• Removing duplicates.
• Handling missing values.
• Converting data types.
• Encoding categorical variables.
• Normalizing or scaling features if needed.

3. Feature Engineering

Relevant software metrics are selected or derived to improve model learning.
Examples may include:

• Complexity metrics.
• Object-oriented metrics.
• Size metrics.
• Coupling metrics.

4. Model Training

Several machine learning models can be trained and compared, such as:

• Logistic Regression.
• Random Forest.
• Decision Tree.
• XGBoost or Gradient Boosting.

5. Model Evaluation

The model is evaluated using metrics such as:

• Accuracy.
• Precision.
• Recall.
• F1-score.
• Confusion Matrix.

6. Model Saving

The best-performing model is saved for future prediction and reuse.

---

Tools and Technologies

• Python
• Jupyter Notebook
• Pandas
• NumPy
• Scikit-learn
• Matplotlib
• Seaborn
• Joblib

---

Installation

Clone the repository:

git clone https://github.com/your-username/code-quality-assessment-model.git
cd code-quality-assessment-model

Create and activate a virtual environment:

python -m venv venv

On Windows:

venv\Scripts\activate

On macOS/Linux:

source venv/bin/activate

Install dependencies:

pip install -r requirements.txt

---

Usage

1. Place the dataset

Put the original dataset in:

data/raw/

2. Run exploration

Open and run:

notebooks/01_data_exploration.ipynb

3. Preprocess the data

Run:

notebooks/02_preprocessing.ipynb

4. Train the model

Run:

notebooks/03_model_training.ipynb

5. Evaluate the results

Run:

notebooks/04_model_evaluation.ipynb

6. Make predictions

If a prediction script is included, run:

python src/predict.py

---

Results

The project will generate:

• Cleaned and processed data.
• Trained machine learning model.
• Evaluation metrics.
• Visualizations of performance.
• Feature importance analysis.

Results will be stored in the reports/ and models/ folders.

---

Future Improvements

• Add more datasets for better generalization.
• Extend the model to support multiple programming languages.
• Improve feature engineering with static analysis tools.
• Build a web interface for real-time code quality prediction.
• Deploy the model as an API.

---

Repository Purpose

This repository is part of my AI/ML portfolio and demonstrates practical skills in:

• Machine learning.
• Data preprocessing.
• Model evaluation.
• Software quality analysis.
• Reproducible project organization.

---

Author

Henikaja Andriamahay IRIMANANA