Rainfall Prediction based on Time-series Data and Temporal Convolutional Network for Flood Forecasting in Malaysia

Extreme rainfall can lead to flood disasters that bring detrimental impacts on various aspects of human lives. Rainfall forecasting is a sophisticated time-series problem due to its nonlinear nature. Recently, deep learning algorithms, such as Long Short-Term Memory (LSTM), have appeared as effective approaches to handling time-series data. However, LSTM has limitations in dealing with long-distance dependencies. Therefore, this study aims to present a deep learning model based on Temporal Convolutional Network (TCN) to address the problem of LSTM in forecasting rainfall amounts for flood prediction. The proposed model was trained and tested using observed rainfall data from multiple states in Malaysia. The performance of the proposed solution was benchmarked with various baseline architectures, including Recurrent Neural Network (RNN), Gated Recurrent Unit (GRU), Bidirectional GRU (BiGRU), Long Short-Term Memory (LSTM), and Bidirectional LSTM (BiLSTM). Several evaluation metrics were used to assess the performance of the proposed model, such as Mean Square Error (MSE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and coefficient of determination ($\mathbf{R}^{2}$). Results from the comparative experiments indicated that TCN outperformed other deep learning models, achieving the lowest MSE, RMSE, and MAE scores of 0.0050, 0.0706, and 0.1467, respectively. The model also obtained the highest $R^{2}$ value of 0.8592, indicating that TCN can effectively handle time-series data and provide accurate prediction of rainfall amounts.