Application of artificial neural network in medical geochemistry

Presented study deals with application of artificial neural network (ANN) in geochemical and health data analysis to define influence of chemical elements in drinking water on health status of population in the Slovak Republic... [ view full abstract ]

Presented study deals with application of artificial neural network (ANN) in geochemical and health data analysis to define influence of chemical elements in drinking water on health status of population in the Slovak Republic (elaborated within the project Geohealth). The first step includes calculation of spatially smoothed values of health indicators through empirical Bayes smoothing. Next step consist of ANN calculations to identify those chemical elements in drinking water that have influence on evaluated health indicators. Influence of particular elements is defined based on values of coefficient of sensitivity (sr). The influence increases with the increase of sr value. Very important task is to identify the optimal number of calculated ANN. ANN with the highest correlation coefficients may not be the most appropriate. They can be wavy with several peaks. That is why in further analysis we use median value from 50 the best ANN with the highest correlation coefficients (from the total number 200 ANN). The results of ANN calculations were verified by methods of non-linear regression, e.g. method of decils. For influential elements is compliance excellent. Limit values of chemical elements for particular health indicators represent intersection of model curve for chemical contents with mean value of health indicator. We define limit (critical) and optimal concentration levels at which level of health indicator is the most favourable.
The project is supported by the EU Life+ programme (LIFE10 ENV/SK/086) and Ministry of the Environment of the Slovak Republic.