Forecaster: Forecast Oriented Feature Elimination-based Classification of Adverse Spatio-Temporal Extremes
Year: 2011
Type: report
Abstract: Accurate forecasting of extreme events in spatio-temporal systems is a highly underdetermined, yet very important problem.Physicsbased models provide unreliable predictions for variables highly related to adverse extreme events, while statistical methods can often only predict linear system responses.In this paper, we propose Forecaster, an algorithm that constructs a forecast-oriented feature elimination-based ensemble of classifiers for robust forecasting of extreme events.In contrast to existing classification methods that predict the current system state, Forecaster predicts the future phase of the system based on the preceding multivariate data, and it is able to handle underdetermined problems.Forecaster supports nonlinear system behavior as well.Experimental results show that Forecaster increases prediction accuracy of traditional methods by up to 13% on two seasonal tropical cyclone prediction systems.
Language: English
Subfield: Atmospheric Science
Field: Earth and Planetary Sciences
Domain: Physical Sciences
SDG: Climate action
Open Access status: green
OpenAlex ID: W7010051432
