The importance of forecasting has grown during the last three decades. This study considers a comparison between actual and forecast by modeling in the context of linear regression model through Box-Cox transformation. The models are fitted with the assumption of non-constant variance of the residuals. First of all, the transformation parameters, the regression parameters and the variance parameters by GARCH process, have been estimated simultaneously using the maximum likelihood method under the classical approach. Secondly, the transformation parameters, and the variance parameters by GARCH process have been estimated under the classical approach while the regression parameters have been estimated under the Bayesian approach using the results of Kalman Filter, the objective of this paper is to reduce the heteroscedastic features in the mean part of the model for better forecast. For this purpose we consider, the value and quantity of rice exported by Pakistan.