This paper addresses the problem of feature selection aiming to improve a flood forecasting model. The proposed model is carried out through a case study that uses 18 different time series of thirty-five years of hydrological data, forecasting the level of the Xingu River, in the Amazon rainforest in Brazil. We employ a Genetic Algorithm for the task of feature selection and exploit several different genetic parameters seeking to improve the accuracy of the prediction. The features selected by the Genetic Algorithm are used as input of a Linear Regression model that performs the forecasting. A statistical analysis verifies that the final model can predict the river level with high accuracy, which obtains a coefficient of determination equal to 0.988. Hence, the proposed Genetic Algorithm showed to be successful in selecting the most relevant features.

本文针对旨在改进洪水预报模型的特征选择问题进行了探讨。拟议的模型是通过一个案例研究进行的，该案例使用了三十五年水文数据的18个不同时间序列，预测了巴西亚马逊雨林中的新谷河水位。我们采用遗传算法来进行特征选择，并利用几种不同的遗传参数来提高预测的准确性。遗传算法选择的特征用作执行预测的线性回归模型的输入。统计分析证明最终模型可以高精度地预测河流水位，从而得出等于0.988的确定系数。因此，