Discovering Causal Hidden Variables Of Bayesian Networks And Its Application

Hidden variables are unobservable, and contain important information about the inherent nature of things. Hidden variables can simplify the structure and converge on the dependencies between variables. Discovering hidden variables is advantageous to realize the true state of things and features for the people. Learning hidden variables of Bayesian network is an important research topic in the field of data mining and knowledge discovery. Discovering hidden variables based on causal analysis and uncertainty analysis, and carry on application of hidden variables from energy based on complex system. The research work of the project has a great practical significance and application value. The research work as follows:Firstly, hidden variable discovering algorithm of structural analysis is difficult to discover hidden variables effectively and possesses of poor interpretability. a learning algorithm of hidden variables is presented based on local causality analysis (LCAHD), Firstly The LCAHD algorithm first obtaining the Markov blanket of interested variable to extract the local dependency structure, then, utilizes interventional to generate interventional data, joint interventional data and observational data to study local causality in the local dependency structure, and then, utilizes causal structure entropy to measure the uncertainty of causality in the local causal structure,and utilizes judging criteria of hidden variables and uncertainty of causality to determine the existence of hidden variables, Finally, hidden variables found algorithm is given.Secondly, the characteristics of the data of stock market are mass and multi-source, which make the data high dimension and lead to lower accuracy of prediction. For this defect, a learning algorithm of hidden variables is presented based on feature fusion, and then use it in stock market research (LHFF). The basic idea of the algorithm as follows:the first is to collect characteristics of stock market and compute correlation between characteristics based on mutual information, and then extract the features based on the value of degree of association. The second is looking the characteristics as hidden variables which carry out feature fusion after be weighted, it is the energy of stock market so that can construct the energy calculation model. The last is using the energy calculation model to get obtain the value of the market energy to predict the Market index. The practical stock market cases show that the algorithm has strong practicability.Experiments on the standard network and stock network, the results of experiment show that this algorithm can effectively determine the location of hidden variables with strong interpretability. The experimental results of financial network show hidden variables have widely applied in real life.