Performance Evaluation of XGBoost and Random Forest Models in Visibility Prediction at Juanda Airport

Abstract

Predicting meteorological visibility is critical in enabling transportation safety and weather watch systems. The current study compares the accuracy of the Random Forest and XGBoost algorithms in performing time series prediction for visibility based on BMKG Juanda, Sidoarjo hourly observation records for one year. Process analysis only uses visibility as a primary variable. Preprocessing of the data involved handling missing values, normalization, and dividing the data into training and test datasets. Model training and hyperparameter tuning were followed by model evaluation using a combination of MAE, RMSE, and MAPE indices. From the results, it is found that Random Forest had an MAE of 808.54, RMSE of 1,312.64, MAPE of 21.09%, and a computation time of 1.02 seconds, while XGBoost had an MAE of 808.81, RMSE of 1,323.12, MAPE of 21.47%, and a computation time of 1.36 seconds. As such, Random Forest is proposed as a more efficient model for predicting visibility at Surabaya's Juanda Airport; however, XGBoost remains a consideration for applicability when there is excessive variability in the data.