Multiclass Prediction of Bug Resolution Time Using Context-Aware Machine Learning Models

Accurately predicting bug resolution time is a persistent challenge in software development and is often complicated by subjective factors and noisy data. This work proposes a robust machine learning pipeline that classifies bug resolution time into four ordinal categories (Immediate, Fast, Normal, and Long) by leveraging project metadata from large-scale repositories, such as GitHub and Jira. The methodology includes comprehensive exploratory data analysis, statistical outlier removal, and advanced feature engineering to derive contextual variables that encapsulate historical projects and component-level behavior. Three machine learning classifiers (Random Forest, XGBoost, and MLPClassifier) were evaluated under various data balancing strategies. While tree-based models showed signs of overfitting, the MLP Classifier achieved superior generalization when enhanced with engineered features, which was validated through cross-validated learning curves and hyperparameter optimization. Our findings underscore the importance of contextual feature design and advance state-of-the-art bug resolution modeling by integrating multiclass classification, temporal context, and model interpretability.