Comparing Linear Regression Methods and Artificial Neural Network in Forecasting Human Mortality as a Function of Air Temperature: Case Study of Tehran City

Seasonal and daily human mortality changes have correlation with air temperature. In this research, daily human mortality data and air temperature during 2002- 2005 has been used. For data analysis, Pearson adjusted correlation coefficient, polynomial regression as a semi-linear method and artificial neural network as a non-linear method have been used. The results of Pearson correlation analysis showed significant negative correlation between air temperature and total human mortality and mortality caused by cardiovascular diseases. Their correlation by artificial neural network and genetic algorithm indicated a better result compared to the classic methods (linear and polynomial regression). After network training with different hidden layers and different stepsizes, it was indicated that the use of artificial neural network with one hidden layer of perceptron results in a better model, in the setting of arranged samples. Therefore, it can be said that neural network can forecast the nonlinear relation between monthly mortality and air temperature, while the combined model of neural network with genetic algorithms can increase analysis speed and accuracy and therefore decrease errors in calculations.