An Algorithm for Bayesian Networks Structure Learning Based on Simulated Annealing with MDL Restriction
YE Shui-sheng, Hong Cai, Rongguan Sun · 2008
By means of deducing and analyzing the minimum description length (MDL) principle as the grading functions, this paper designs a maximum entropy grade-function with complexity restriction, and proposes an algorithm for structure learning in Bayesian networks based on simulated annealing. Then according to the analysis of the historical stock data with this algorithm, the topological structure of network is obtained and so is conditional probability table of every network node. At last, the trend and fluctuation interval of the stock price are forecasted by this Bayesian model.