Sparse Bayesian Flood Forecasting Model Based on SMOTEBoost

Flood is a common disaster in our daily life. It's of great significance to improve the accuracy of flood forecasting, in order to help get rid of loss in both lives and property. However, there exists a uneven distribution of samples in factors of flood forecasting. Therefore, it's difficult to train a single data driven model to describe the entire complex process of flood generation. In this paper, we propose a novel SMOTEBoost algorithm to perform flood forecasting with both high accuracy and robustness. Specifically, we firstly adopt a SMOTE algorithm to generate virtual samples, which greatly alleviates the problem of uneven sample distribution. Afterwards, we propose a sparse Bayesian model, which is trained with AdaBoost training strategy by improving its performance in over-fitting. At last, we carry out experiments on flood foretasting in Changhua river, which shows that the proposed method achieves high accuracy in prediction, thus owing practical usage.