TOWARDS A LEARNING MODEL BASED ON BAYESIAN NETWORKS