Bayesian Networks Parameter Learning Based on Noise Data Smoothing in Missing Information

A parameter learning algorithm based on noise data smoothing is developed in static Bayesian Networks (BN) to tackle the problem of randomly missing observed information, i.e., data missing can occur arbitrarily in every group of data in the sample. the simulation results demonstrate that this algorithm has similar speed and accuracy compared with EM algorithm in the condition of missing proportion less than 20 percent. the parameter learning precision is better than EM algorithm (more than 20% missing data).