Enhancing the Accuracy of Airline Review Classification Using SMOTE and Grid Search with Cross Validation for Hyperparameter Tuning

Abstract

This research aims to improve the classification accuracy of online airline reviews using SMOTE (Synthetic Minority Over-sampling Technique) and Grid Search with Cross Validation (CV) for hyperparameter tuning. The dataset includes 23,170 airline reviews processed through preprocessing, feature extraction, and data sharing stages. Ensemble models such as, LightGBM, XGBoost, GBM, and Random Forest were applied. SMOTE handles data imbalance, while Grid Search improves model performance. Results show that the optimized LightGBM achieved the highest accuracy of 99.10%, surpassing other models in precision, recall, and F1-score. This research provides important insights for airlines in understanding customer satisfaction and improving data-driven services.