Influence of Anomalies on the Models for Nitrogen Oxides and Ozone Series

Nowadays, observing, recording, and modeling the dynamics of atmospheric pollutants represent actual study areas given effects pollution on population ecosystems. The existence aberrant values may influence reports air quality when they are based average over a period. This also models, which further used in forecasting. Therefore, correct data collection analysis is necessary before modeling. aimed to detect nitrogen oxide concentration series recorded interval 1 January–8 June 2016 Timisoara, Romania, retrieved from official National Network for Monitoring Air Quality, Romania. Four methods were utilized, including interquartile range (IQR), isolation forest, local outlier factor (LOF) methods, generalized extreme studentized deviate (GESD) test. Autoregressive integrated moving (ARIMA), Generalized Regression Neural Networks (GRNN), hybrid ARIMA-GRNN models built after removal values. results show that first approach provided good model (from statistical viewpoint) anomalies removal. best was obtained by ARIMA-GRNN. For example, raw NO2 series, ARIMA not statistically validated, whereas, without outliers, ARIMA(1,1,1) validated. GRNN able learn well: R2 = 76.135%, correlation between predicted (rap) 0.8778, mean standard errors (MSE) 0.177, absolute error MAE 0.2839, percentage MAPE 9.9786. Still, test set, worse: MSE 1.5101, 0.8175, rap 0.4482. training set better than (R2 0.996), very 0.473). performances ARIMA–GRNN initial satisfactory (the pattern computed almost linear) but outliers close 1). same true O3.