NARX Neural Network for Imputation of Missing Data in Air Pollution Datasets

Abstract:

Air pollutant sensors capture large amounts of data; some of this information is lost due to various causes, including sensor errors and human error. This work proposes a method for the imputation of missing data through a NARX neural network implemented in Matlab. Data pre-processing included standardization of input variables and removal of outliers. Subsequently, the value of the angle between the input variables’ levels was calculated considering 10-min intervals. The neural network uses pollutants O3, CO, NO2, SO2, PM2_5, and temperature as input variables according to the previous analysis of interactions between contaminants. Data indicated that the method obtained the best results for the O3 and NO2 with values of R = 0.85 and R = 0.73, respectively.

Año de publicación:

2020

Keywords:

• NARX neural network
• Data Mining
• Data imputation
• Palabras clave: Neural networks

Fuente:

scopus