Prediction model of missing data: a case study of PM10 across Malaysia Region

Abstract

PM₁₀  is one of the major concerns that have high potential for harmful effects on human health. Thus, prediction of PM₁₀  was performed with the objectives to model suitable PM₁₀ prediction  formula  to  predict  the  concentration  of  PM₁₀. Imputation methods  of  EMB-algorithm and nearest neighbor were applied to treat missing data before analyzed by Fit model, MLR and ANN. R² obtained for Fit-model, MLR and ANN using imputation method of EMB-algorithm and nearest neighbor are (0.9975, 0.3858), (0.9623, 0.3857) and (0.9975, 0.4025) respectively. Sensitivity analysis (SA) shows humidity, temperature, CO, UVB and O₃  out of fifteen parameters contribute the most to the present of PM₁₀ concentration. In conclusion, formula for the best PM₁₀ prediction can be modeled by using ANN or Fit model together with the imputation method of EMB-algorithm.

Keywords: PM₁₀ prediction; fit-model; MLR, ANN; imputation method