Comparison Between Traditional Models and the Use of Artificial Intelligence Models (Neural Networks) to Predict Tobacco Production in Lattakia-Syria

Majd Namaa (1)*

(1). Agriculture Economics., Fac. Agri, Tishreen Univ., Lattakia, Syria.

(*Corresponding author: Dr. Majd Namaa, E-mail:

Received:28/12/2021                         Accepted:14/02/2022


The aim of the research is to compare the traditional predictive models using the multiple regression model, the (ARIMA) model and the neural network model in terms of the predictive ability of tobacco crop production in Lattakia Governorate using some statistical criteria as the mean of the errors of the estimated model and the average differences between the real values ​​and the expected values ​​for each model. The research relied on the data of the annual agricultural statistical group for data on production, productivity, and cultivated area during the period (1991-2019) in addition to data on temperature and annual precipitation, where agricultural production was adopted as a dependent variable (production year, cultivated area, average annual temperature, and average rainfall annual) as independent variables. The stepwise multiple regression method was used to estimate the regression model, the Expert Modeler method was used to estimate the ARIMA model, and the multi-layered module (Perceptron MLP) was used to build the neural network model and test its accuracy, as 23 years of data were used in the training phase. at a rate of (79.3)%, and 6 years for the testing phase with a rate of (20.7%).The results of the research showed the superiority of the neural network model over the regression model and the (ARIMA) model in terms of predictive ability using the mean squares of errors criterion for the estimated model and the criterion of the average differences between real and expected values, where the mean squared error was (1.66) using the neural network model versus (4.47) Using the regression model and (36.123) using the (ARIMA) model, and the average difference between the real and expected values ​​using the neural network was (1426.48) versus (1451.16) for the regression model and (1623.73) for the (ARIMA) model.

key words: production, tobacco, ARIMA, multilayer neural network.

Full paper in Arabic: pdf