Uma avaliação de algoritmos de regressão para predição de volume de chuva
Guilherme S.E. Ferreira, Dianne Medeiros

DOI: 10.14209/sbrt.2022.1570817471
Evento: XL Simpósio Brasileiro de Telecomunicações e Processamento de Sinais (SBrT2022)
Keywords:
Abstract
The occurrence of extreme events that cause natural catastrophes is difficult to predict due to the chaotic nature of these events. A first step to accurately predict these events in time is to build models capable of predicting the rainfall level in a short time-window. This paper evaluates the performance of 10 regression algorithms in order to identify those that generate assertive models to predict the level of precipitation at each hour. We analyze the coefficient of determination and the root mean square error, calculated for each model. The results show that the random forest and degree 3 polynomial regression algorithms achieve the lowest average errors, being potential candidates for rainfall forecasting.

Download