This paper presents a sequence-based activity prediction approach which uses Bayesian networks in a novel two-step process to predict both activities and their corresponding features. In addition to the proposed model, we also present the results of several search and score (S&S) and constraint-based (CB) Bayesian structure learning algorithms. The activity prediction performance of the proposed model is compared with the naïve Bayes and the other aforementionedS&S and CB algorithms. The experimental results are performed on real data collected from a smart home over the period of five months. The results suggest the superior activity prediction accuracy of the proposed network over the resulting networks of the mentioned Bayesian network structure learning algorithms.