Data mining of Iran's 24-hour air temperature by the use of multi-layer perceptron neural network machine learning

The main goal of this research was analyzing the 24-hour air temperature of synoptic weather stations in Iran. The materials and data used in the research were from the hourly data from the National Meteorological Organization for a 31 years period. The data was simulated by the R programming language of the MLP multilayer perceptron neural network. The ANOVA function was used to compare the average 24-hour air temperature in the stations under investigation for further simulation. Using the rotation function, the time patterns in the data were analyzed to in order determine whether the data sequences were random or had significant patterns. In the continuation of the learning methods, logistic regression was applied aiming at the predicting the effects of climate changes in air temperature variations. In the logistic model the climate changes were chosen as the dependent variable and air temperature (observational and simulated) as the independent predictor variables. The data were included in the analysis and the results of applying the logistic model were significant. The Chi square function of the temperature was calculated as 314.19, which was significant at the error level of less than 0.05. The mentioned independent variables were able to correctly explain between 92 and 88 percent of the changes that led to an increase or decrease in air temperature. 86.4% of the months that had no changes were correctly classified, and 93.2% of the predictions about air temperature changes were correct. In general, 95.3% of the predictions were estimated correctly. The results showed that climate changes have a significant effect on increasing or decreasing the monthly air temperature. The simulations predicted the highest and lowest observed temperatures to be in July and January, respectively. The highest and lowest annual air temperatures were recorded and predicted for Siri and Zarineh stations, respectively