Investigating artificial neural networks versus regression models in predicting MI mortality based on climatic elements in Sanandaj

Subject Areas :

Bromand Salahi ¹ , Seyed Asaad Hosseini ² , Kaweh Mohammadpour ³

1 - Professor of climatology, Faculty of Social Science, University of Mohaghegh Ardabili, Ardabil, Iran.
2 - University of Mohaghegh Ardabili, Ardabil, Iran
3 - University of Kharazmi, Tehran, Iran

Received: 2023-06-14 Accepted : 2023-09-11 Published : 2023-09-11

Keywords: Climatic Parameters, Myocardial Infarction, Sanandaj City, Regression Model, ANNs,

Abstract :

To analyze the relationship between mortality due to Myocardial Infarction (MI) with climatic parameters and its prediction, the ability of artificial neural network models, and linear and nonlinear regression in Sanandaj was evaluated. The dependent variable is the total number of MI mortality. 54 months out of 60 in the sample period (2014-2018) were dedicated to training the ANNs model and, the remaining six months were given to test the ANNs model. By a selection of the monthly average temperature, a monthly average of maximum and minimum temperature, the average of the maximum and minimum air pressure measured at earth surface, the total number of sunny hours, and the number of days that their temperature is equal to or less than zero as input, a three-layer perceptron accompanied the didactic Levenberg-Marquardt algorithm and a hidden layer which contained 13 neurons and movable function of sigmoid tangent had the best possible output (the number of MI mortality). The results showed that in the relationship between the monthly MI mortality with the climatic parameters of Sanandaj, The relative error for multiple linear and nonlinear regression models is 22.3% and 22.1%, respectively, while for the ANNs model, it is 2.6%. The results also showed that according to the model error, using ANNs model as a nonlinear method in predicting and diagnosing the relationship between climatic parameters and mortality due to MI in Sanandaj could be considered an efficient and powerful tool in comparison with regression models.