| Summary: | The ultraviolet index is an international standard metric for measuring the strength of the ultraviolet radiation reaching Earth’s surface at a particular time, at a particular place. Major health problems may arise from an overexposure to such radiation, including skin cancer or premature ageing, just to name a few. Hence, the goal of this work is to make use of Deep Learning models to forecast the ultraviolet index at a certain area for future timesteps. With the problem framed as a time series one, candidate models are based on Recurring Neural Networks, a particular class of Artificial Neural Networks that have been shown to produce promising results when handling time series. In particular, candidate models implement Long Short-Term Memory networks, with the models’ input ranging from uni to multi-variate. The used dataset was collected by the authors of this work. On the other hand, the models’ output follows a recursive multi-step approach to forecast several future timesteps. The obtained results strengthen the use of Long Short-Term Memory networks to handle time series problems, with the best candidate model achieving high performance and accuracy for ultraviolet index forecasting.
|
|---|