Table 2 Predictive confusion matrices for each network structure, displaying numbers of true positives, false positives, true negatives, and false negatives. These values were calculated by learning the structure and parameter values from a random subset forming 90% of the overall data, then making predictions for whether a household is food secure or not for the remaining 10% of the data. These show less distinction between the different network structures in terms of performance but using expert knowledge as a prior still seems to give a small boost

From: Bayesian belief network modelling of household food security in rural South Africa