Comparative study of ANN and XGBoost in Water Level Prediction: A Case Study of the Pasak River in Thailand

Abstract

Flooding and drought are some of the natural disasters in Thailand. It is a challenge to manage water resources to control the water level. However, many dams have been built to sufficiently hold Thailand's water resource reservation in the dry season. Oversupply water retention lead to flooding in the rainy season. Accurate forecasting of water level is, therefore, one of the important factors to make a decision-making process in controlling water level. This research aims to study machine learning techniques for water level prediction in the Pasak River basin. Extreme Gradient Boosting, a useful machine learning technique, has been implemented. Water level and precipitation of stations PAS001, PAS002, PAS003, and PAS004 are variables, and water level in PAS004 is the target prediction. The study found that ANN and XGBoost algorithm are accurate in predicting water levels. The evaluation means absolute error (MAE) by ANN is 2.42 cm. and XGBoost is 3.14 cm. The ANN algorithm uses computer’s memory much more and takes time for learning process longer than XGBoost. Keywords: Machine learning, Water level prediction, Artificial Neuron Network, XGBoost