Artificial neural network modeling and exposure assessments: a new scaling approach

Abstract Conventional modeling approach such as regression has not much ability to incorporate nonlinearity of environmental covariates during different categories of estimations. Nonlinear information is considered as outliers in prediction while fitting in the regression models. The ANN models are robust due to adaptive neuro computing to consider these kinds of parameters in the scaling and instrumentation process due to gradient neural weight training mechanism. Raised level of ambient particulate matter having size PM10 and PM2.5 is a serious matter of concern for children having age group 8 to 14 years. A study has been conducted in North West side of Asia continent for three years (2013 to 2016) where sampling was done to measure ambient PM level, physiological parameters of 600 children and covariate data. A hybrid recurrent neural network (HRNN) has been developed for exposure assessment with the help of Gamma Test. Based on Root Mean Square Error and Mean Absolute Percentage Error scale, the trends in physiological parameters were estimated. The developed model strategy, will help research community and policy-makers to feed time series data in the model for spatiotemporal analysis rather than conventional statistical methods.