Publicación: Interpretable forecasts of NO2 concentrations through deep SHAP
Cargando...
Fecha
2019-09-25
Autores
Editor/a
Director/a
Tutor/a
Coordinador/a
Prologuista
Revisor/a
Ilustrador/a
Derechos de acceso
Atribución-NoComercial-SinDerivadas 4.0 Internacional
info:eu-repo/semantics/openAccess
info:eu-repo/semantics/openAccess
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Nacional de Educación a Distancia (España). Escuela Técnica Superior de Ingeniería Informática. Departamento de Inteligencia Artificial
Resumen
Increasing in pollution levels in cities, especially in developed countries, and the consequences that it has on health and environmental, have prompted institutions to take preventive measures to reduce pollution levels. European Union has established thresholds for certain gases, such as nitrogen dioxide (NO2). If these thresholds are exceeded, institutions of each city belonging to the EU, must activate measures previously defined, to go down NO2 concentrations. Meteorological parameters are related to episodes of high/low NO2 concentrations. In this paper, pollution time series and meteorological features measured in the Madrid region were used to predict NO2 concentrations. Detailed relationships between NO2 concentrations and meteorological data were established through computational intelligence models. Therefore, these data were used to develop a predictive model. The proposal shows good precision in the prediction, proving computational intelligence has great potential in the pollution time series forecasting. When constructing a prediction model, in addition obtaining good accuracy, it is important to know why that prediction is made. Interpreting the output of the computational intelligence model is difficult due to the architecture of the model itself. A SHAP (SHapley Additive Explanations) approach is applied to interpret the complex model outputs. This method assigns each feature a value of importance for a particular prediction. Three SHAP-based explanation methods are compared to determine which method is more suitable for the pollution time series data and for the computational intelligence model chosen. In this research therefore, a model based on computational intelligence is developed to predict the levels of NO2 concentrations. Through methods based on explanations we will obtain a deeper vision of how the computational intelligence model behaves on the pollution time series, allowing an increase in the confidence of the users on the results obtained from the prediction model. As a product of this study, a short version of this document has been sent for consideration in the Ecological Informatics1 international journal on computational ecology and ecological data science.
Descripción
Categorías UNESCO
Palabras clave
Citación
Centro
Facultades y escuelas::E.T.S. de Ingeniería Informática
Departamento
Inteligencia Artificial