Bayesian Decision Tree Averaging for the Probabilistic Interpretation of Solar Flare Occurrences

Bayesian averaging over Decision Trees (DTs) allows the class posterior probabilities to be estimated, while the DT models are understandable for domain experts. The use of Markov Chain Monte Carlo (MCMC) technique of stochastic approximation makes the Bayesian DT averaging feasible. In this paper we describe a new Bayesian MCMC technique exploiting a sweeping strategy allowing the posterior distribution to be estimated accurately under a lack of prior information. In our experiments with the solar flares data, this technique has revealed a better performance than that obtained with the standard Bayesian DT technique.