Enhancing sediment transport predictions through machine learning-based multi-scenario regression models

Abstract

Machine learning is one effective way of increasing the accuracy of sediment transport models. Machine learning captures patterns in the sequence of both structured and unstructured data and uses it for forecasting. In this research, the different regression models were train to forecast sediment data using 8 years of measured sediment data collected in Sg. Linggui suspended sediment station. Data from different scenarios were used where each scenario indicates the number of lags. Seven regression models, namely, Linear Regression, Regression Trees, Support Vector Machines, Gaussian Process Regression, Kernel Approximation, Ensemble of Trees, and Neural Network were trained using the data and compared. The trained models were evaluated using Root Mean Square Error (RMSE) and Coefficient of Determination (R2). The best-performing models from two different types of regression models were chosen and they were tested using the test data to find the Relative Percentage Error (RPE) of the predicted data. The Exponential Gaussian Process Regression model performs much better than the other models in terms of RMSE and R2 values. When the exponential models from all 3 scenarios are compared, scenario 3 seems to have a better-performing model but only by a very small margin, after using testing data, the result shows scenario 3 has less RPE compared to other models. Hence, it can be deduced that the exponential gaussian process regression model from scenario 3 is the best-performing model overall in terms of RSME, R2, and RPE.