Bug Fixing Time Prediction using ALBERT, DistilBERT, and Google BERT

This repository contains the implementation for predicting bug-fixing time using transformer-based language models. The models evaluated include ALBERT, DistilBERT, and Google BERT. The dataset consists of bug reports extracted from LiveCode's Bugzilla, with textual features such as bug descriptions and developer comments used for training.

Features

• 📜 Data Processing: Extracts and preprocesses bug reports from a JSON dataset.
• 🔍 Model Comparison: Evaluates ALBERT, DistilBERT, and Google BERT for bug-fixing time prediction.
• 📊 Performance Metrics: Computes Accuracy, F1-score, RMSE, and Inference Time.
• 📈 Visualization: Generates comparative plots for model performance analysis.

Installation

To run this project in Google Colab, ensure the required dependencies are installed:

pip install torch transformers scikit-learn matplotlib

Usage

Upload your dataset (LiveCode_original.json) and run:

python bug_fixing_time_prediction.py

Dataset

• The dataset is extracted from LiveCode's Bugzilla.
• Each bug report includes the short description, developer comments, and resolution time (days_resolution).

Results

The models are evaluated based on:

• Accuracy: Correctness of predictions.
• F1-score: Balance between precision and recall.
• RMSE: Measures deviation between predicted and actual bug-fixing time.
• Inference Time: Measures model efficiency.