Using Empirical Covariance Matrix in Enhancing Prediction Accuracy of Linear Models with Missing Information