Petar Sekulic, Paola Regina, Luana Spadafina, Giuseppe Dentamaro, Alessandro Porcelli, Cristiano Bove, Slavko Kovacevic, Mirko Kalezic
Real-time flood prediction using Recurrent Neural Networks and Random Forest
Floods are one of the most destructive natural disasters as they cause severe material damage and often the loss of human life. Predicting a flood is a challenging task and recent progress in this field was brought by machine learning (ML) models. This paper aims to discover a dependency between spatial and temporal data patterns about the ground and weather that will subsequently lead to the floods. Taking into account features deriving from remote sensing images and weather stations, the proposed algorithm aims to predict whether the flood will happen or not by using Recurrent Neural Networks and Random Forests methods. The algorithm could be considered as a starting point of the research related to prediction of other natural disasters, such as landslides, heavy rainfall, droughts, weather forecasting, etc. The future direction of research aims at improving the accuracy of the algorithm by employing a broader and more structured data set built from validated sources.