Optimal tuning parameter estimation in maximum penalized likelihood method