Short term forecasting is significant operation to forecast the future jobs for computational grids as it can provide a solution for inconsistent resource availability and feasible job scheduling. A job forecasting model is presented to forecast one hour ahead of jobs submitted for computations using regression random forests. The training data constitutes the information about the type of job and jobs submitted on average each hour. The forecast model is built on the basis of training process. A real job data set from LCG (Large Hadron Collider Computing Grid) is used for evaluating the proposed forecast model, while considering the fact that jobs submitted are inconsistent. Findings provide a proof that by using proposed method the forecast error can be reduced and the effectiveness of job forecast can be improved for long test periods. © 2016 IEEE.