Comparing Neural Network and Multiple Regressions Models to Estimate Monthly Rainfall Data

Marathada region situated between 170 -35 N and 200- 40 N latitude and 740 - 40 E and 780 15 E longitudes in Maharashtra state. In this region consist of seven districts such as, Beed, Hingoli, Jalna, Latur, Nanded, Osmanabad and Parbhani. The monthly data of rainfall, maximum & minimum temperature, minimum & maximum humidity, wind speed, wind direction & total could cover (octa) of five years (2 009 to2014). For analysis of data, statistical methods based on multiple regression and artificial neural networks have been used. According to results of this paper, a comparison between multiple regression and neural network models, using the SPSS & Mat lab software. Two statistical parameter multiple correlation coefficient (R) and Mean Square Errors (MSE) is used for selection best fit model. The model have maximum correlation coefficient (R) and minimum Mean Square Errors (MSE) that model is best fit to estimate the monthly rainfall of Marathwda region. Result show that Aurangabad station value of the Multiple correlation coefficient (R) of Aurangabad satiation 0.915 for multiple regression model while it was 0.949 in neural network model, secondly, the mean square error (MSE) for multiple regression model is 2.41 which is higher than that of the mean square error (MSE) obtained through neural network model as 0.945 on comparing of these result finally concluded the Aurangabad station have Artificial Neural Network (ANN) is best model for estimation of monthly rainfall data. Finally on the basis of maximum value of Multiple correlation coefficient (R) and minimum value of mean square error (MSE). it was concluded that the selected metrological station neural network model is better than multiple regression model for estimation of monthly rainfall data