Classification of Refactoring types using Commit messages

This repository contains code and data for a multi-label classification model for refactoring types based on the commit messages. The model uses machine learning algorithms to predict the types of refactoring that may be appropriate for a given dataset of commit messages. Six method-level Refactoring types considered were: Extract Method, Inline Method, Move Method, Pull-up Method, Push-down, Rename.

Installation

To install the necessary packages, we added a code cell in the beginning of each script with the install commands. No external installation required.

Usage

The Classification_Of_Refactoring_Types.ipynb script can be used to train and test the model on a given dataset for multi class classification (Phase 2). The FeatureCount.ipynb script can be used to get the useful feaures in the commit messages for each reafactoring type. The Multi-Label_Refactoring_Types.ipynb script can be used to train on different models and predict the refactoring types to get the best performing model.

Datasets

Datasets used for this project are present in the file Data

• Message.csv is the dataset provided to us for analysis with commit meassges and class ( refactoring types)
• Xtest.csv is the test data with commit meassages after pre-processing for testing the model
• y_pred_gb.csv is the predictions data of gradient boost for feature count in mutli-class classification
• y_pred_lr.csv is the predictions data of logistic regression for feature count in mutli-class classification
• y_pred_rf.csv is the predictions data of random forest for feature count in mutli-class classification
• y_test.csv is the test data with classes for testing the model

Scripts:

1. Classification_Of_Refactoring_Types.ipynb script has be used to train the model on a given dataset. (phase 2) The script requires Message.csv dataset

   • Model

     • Random Forest
     • Gradient Boosting
     • Logistic Regression
     • Support Vector Classification
     • Naive Bayes

   • Evaluation

     The model's performance can be evaluated using precision, recall, and F1-score.

2. FeatureCount.ipynb script requires the below datasets (phase 2)

   • Xtest.csv
   • y_pred_gb.csv
   • y_pred_lr.csv
   • y_pred_rf.csv
   • y_test.csv

3. Multi-Label_Refactoring_Types.ipynb The script requires Message.csv dataset (phase 3)

   • Model

     We have used 3 different approaches namely Label Powerset, Binary Relevance and Classifier Chains.

     • Multinomial Naive Bayes
     • Stochastic Gradient Descent
     • Logistic Regression
     • Gaussian Naive Bayes
     • Decision Tree
     • Random Forest

   • Evaluation

     The model's performance can be evaluated using micro and macro average values for precision, recall, and F1-score. Also Subset accuracy and Hamming loss can be used for the multi-label classification.