This study has been conducted using the quantities measured by the Mehrabad Airport Automatic Station Sensors during January to May 2013. During this period, the largest number of Gusty wind occurred at the station. Results for Sensor of Band No. 29 are presented here. All data, except the data of Gusty wind direction and speed have been normalized to interval 0.1-0.9. Then 70% of all data were selected for training, 15% for test and 15% for validation.
Measured quantities were processed using the sequential forward and backward feature selection, and mutual information (MI) with the maximum-relevance minimum-redundancy criterion. Then, the selected features were used in linear regression to predict Gusty wind direction and velocity.
The result shows that the selected features form the sequential backward method for velocity of gusty wind is better than the other two methods. Among the months, the selection features for February have fewer errors. Comparing the two seasons, the results of winter are better than spring. But the selected features for the Gusty wind direction show that the error rates of all three methods are approximately the same.