Flood prediction of Sungai Bedup, Serian, Sarawak, Malaysia using deep learning

Abstract

This paper aims to evaluate the performance of the Long Short Term Memory (LSTM) model for flood forecasting. Seven data sets provided by the Drainage and Irrigation Department (DID) for Sungai Bedup, Serian, Sarawak, Malaysia are used for evaluating the performance of LSTM algorithm. Distinctive network was trained and tested using daily data obtained from the DID with the year range from 2014 to 2017. The performance of the algorithm was evaluated based on (Training Error Rate, Testing Error Rate, Loss, Accuracy, Validate Loss and Validate Accuracy) and compared with the Backpropagation Network (BP). Among the seven data sets, Sungai Bedup showed small testing error rate which is (0.08), followed by Bukit Matuh (0.11), Sungai Teb (0.14), Sungai Merang (0.15), Sungai Meringgu (0.12), Semuja Nonok (0.14) and lastly Sungai Busit is (0.13). Moreover, the developed model performance is evaluated by comparing with BP model. Results from this research evidently proved LSTM models is reliable to forecasting flood with the lowest testing error rate which is (0.08) and highest validate accuracy (92.61% ) compared to BP with testing error rate (0.711) and validate accuracy (85.00%). Discussion is provided to prove the effectiveness of the model in forecasting flood problems.