A desktop grid is heterogeneous collections of local and volunteer resources. These resources can be assigned to heterogeneous jobs whereas these resources cannot be guaranteed to be available every time of job execution. Therefore, the resource availability and load forecast can help to minimize the job failures and job migration. In this paper, a forecast based proactive job scheduling and migration (PJS-ANN) has been proposed using artificial neural networks to make load forecasts for scheduling the jobs to reliable volunteer resources. The proposed method performance has been compared with conventional load balancing (LB) and no-migration (NM) algorithms. The performance comparisons demonstrate that the PJS-ANN has lower turnaround time per job and job failure rate has been significantly improved.