Identification of significant factors for air pollution levels using a neural network based knowledge discovery system

Artificial neural network (ANN) is a commonly used approach to estimate or forecast air pollution levels, which are usually assessed by the concentrations of air contaminants such as nitrogen dioxide, sulfur dioxide, carbon monoxide, ozone, and suspended particulate matters (PMs) in the atmosphere of the concerned areas. Even through ANN can accurately estimate air pollution levels they are numerical enigmas and unable to provide explicit knowledge of air pollution levels by air pollution factors (e.g. traffic and meteorological factors). This paper proposed a neural network based knowledge discovery system aimed at overcoming this limitation in ANN. The system consists of two units: (a) an ANN unit, which is used to estimate the air pollution levels based on relevant air pollution factors; (b) a knowledge discovery unit, which is used to extract explicit knowledge from the ANN unit. To demonstrate the practicability of this neural network based knowledge discovery system, numerical data on mass concentrations of PM2.5 and PM1.0, meteorological and traffic data measured near a busy traffic road in Hangzhou city were applied to investigate the air pollution levels and the potential air pollution factors that may impact on the concentrations of these PMs. Results suggest that the proposed neural network based knowledge discovery system can accurately estimate air pollution levels and identify significant factors that have impact on air pollution levels.

► Common ANN models cannot provide explicit knowledge on significant contributors to air pollutants.
► ANN based knowledge discovery system was proposed to overcome this limitation.
► This new system can produce accurate estimates on air pollutants levels (PM2.5 and PM1).
► This new system can also identify significant contributors to PM2.5 and PM1.