A Feature Selection Method for Air Quality Forecasting

Abstract. Local air quality forecasting can be made on the basis of meteorological and air pollution time series. Such data contain redundant information. Partial mutual information criterion is used to select the regressors which carry the maximal non redundant information to be used to build a prediction model. An application is shown regarding the forecast of PM10 concentration with one day of advance, based on the selected features feeding an artificial neural network.

• Type: Conference
• Year: 2010
• Where: ICANN
• Authors: Luca Mesin, Fiammetta Orione, Riccardo Taormina, Eros Pasero
• Keywords: Air Quality Forecasting, ICANN 2010, Maximal Non Redundant, Neural Networks, Redundant Information