Exploiting statistical correlations for proactive prediction of program behaviors