Forecasting Water Levels Using Data Mining Techniques : A case study at Nong Han Lake, Sakon Nakhon, Thailand

Abstract

Nong Han is the largest lake in the northeast region of Thailand. Looking at the flood crisis in the year 2017, Sakon Nakhon Province was a greatly affected area. This research aims to compare the efficiency of data mining techniques for predicting changes in water level in Nong Han by three methods, as follows: 1) Support Vector Machine, 2) Linear Regression, and 3) Multi-Layer Percepton to create a suitable model for predicting changes in water level in Nong Han. Mean absolute error, was employed to assess the model's efficiency for selecting the best method to develop the forecasting model through analysing daily data collected during the years 2011 – 2016, totaling 1,961 items.

Comparing the efficiency of the different water level forecasting models for the Nong Han Lake, it was found that the Multi-Layer Perceptron was the highest performing model with a mean absolute error of 7.921. Next, the Linear Regression with a mean absolute error of 8.343. Finally, the Support Vector Machine with a mean absolute error of 10.824.