Performance of Water Level Forecasting Based on Chaos Approach Using Data Splitting

Abstract

Forecasting accuracy should be prioritised in flood plain areas. This research focuses on data split of water level time series datasets in producing excellent forecasts as measured by coefficient correlation (CC). The datasets involved 6000 hours chosen from a recent research location at Sungai Dungun water level, which the data was split into different ratio datasets (50:50, 60:40, 70:30, 80:20, 90:10). A recent study has proved that the chaotic dynamic existed in time series data when running the data using the Cao method. The dataset used was divided into training and testing data to evaluate the performance based on the local linear approximation method. Those sets of data required a combination of parameters for prediction. In this study, the data split of water level time series data gave impacts to the combination of parameters for prediction. The result obtained was in the range of strong forecast using chaos approach with over 95% accuracy in every dataset. In addition, the dataset with a 50:50 ratio showed the highest CC obtained, and its values decreased in ascending order of 60:40, 70:30, 80:20, and 90:10. It showed that the splitting data of training and testing had an impact on prediction results. The higher number of training data ran, the lower number of CC was obtained. However, the chaos method still gives excellent prediction results, even when forecasting using different ratios of data set.