Application of machine learning in predicting salinity in Tran De estuary

The study aims to apply machine learning methods to predict salinity in the Dai Ngai hydrological station area. The input data collected over a period of 10 years includes Tran De water level, Tran De salinity, Can Tho flow, and the predicted parameter is Dai Ngai salinity. These parameters after being pre-processed, are used to train ANN, RF, and XGBoost models based on 5 scenarios built through Spearman correlation analysis results. The evaluation results showed that the ANN model trained on scenario 5 achieves the best performance with R 2 value obtain 0.688, while MAE and RMSE are very low and predict relatively accurate the salinity cycle at the Dai Ngai hydrological station. The results of the study also revealed the applicability of machine learning models in the situation of salinity intrusion in the downstream of the Hau River, which has fluctuated erratically in recent years.