Comparison of Random Forest, SVR and KNN Based Models in Sea Level Prediction for Erdemli Coast of Mersin

Abstract

Seawater level prediction is very important in terms of future planning of human living conditions, flood prevention and coastal construction. Nevertheless, it is hard to correctly predict the daily future of sea water level because of the atmospheric conditions and effects. Therefore, Random Forest (RF), Support Vector Regression (SVR) and K-Nearest Neighbor (KNN) methods were used for the prediction of seawater level on Erdemli coast of Mersin in this study. In this paper, root mean square error (RMSE) and coefficient of determination (R2) were applied as model evaluation criteria. In addition, 15-minute sea water level data of Erdemli Station for approximately 18 months were obtained and used as is. The results depict that Random Forest model can predict the seawater level for 1st and 2nd days with R2 of 0.80, 0.63, respectively, KNN model can predict for 1st and 2nd days with R2 of 0.80, 0.64, respectively, and SVR model can predict for 1st and 2nd days with R2 of 0.77, 0.60, respectively.