USE OF MULTIVARIATE MACHINE LEARNING ANALYSIS TECHNIQUES FOR FLOOD RISK PREVENTION
- Politecmico di Milano, Department of Electronic Information and Bioengineering, Milan, Italy
Keywords: flood, risk, data, Machine Learning, Prediction
Abstract. Natural disasters such as flood are regarded to be caused by extreme weather conditions as well as changes in global and regional climate.
The prediction of flood incoming is a key factor to ensure civil protection in case of emergency and to provide effective early warning system. The risk of flood is affected by several factors such as land use, meteorological events, hydrology and the topology of the land.
Predict such a risk implies the use of data coming from different sources such satellite images, water basin levels, meteorological and GIS data, that nowadays are easily produced by the availability new satellite portals as SENTINEL and distributed sensor networks on the field.
In order to have a comprehensive and accurate prediction of flood risk is essential to perform a selective and multivariate analyses among the different types of inputs.
Multivariate Analysis refers to all statistical techniques that simultaneously analyse multiple variables.
Among multivariate analyses, Machine learning to provide increasing levels of accuracy precision and efficiency by discovering patterns in large and heterogeneous input datasets.
Basically, machine learning algorithms automatically acquire experience information from data.
This is done by the process of learning, by which the algorithm can generalize beyond the examples given by training data in input. Machine learning is interesting for predictions because it adapts the resolution strategies to the features of the data. This peculiarity can be used to predict extreme from high variable data, as in the case of floods.
This work propose strategies and case studies on the application on machine learning algorithms on floods events prediction.
Particullarly the study will focus on the application of Support Vector Machines and Artificial Neural Networks on a multivariate set of data related to river Seveso, in order to propose a more general framework from the case study.