Efficient Parameter Estimation for Bayesian Network Classifiers using Hierarchical Linear Smoothing