Research And Application On Bayesian Networks In Intelligence Data Processing

A Bayesian network is a graphical model for probabilistic relationships among a set of variables. This provides a natural tool for dealing with two problems, uncertainty and complexity, in applying mathematics and engineering. Over the last two decade, the Bayesian network has become a popuiar representation for encoding uncertain expert knowledge in expert systems. With the expansion of database scales, Bayesian networks have been applied in large-scales database for data mining and knowledge discovery and become a powerful tool for decision-making. Bayesian networks play an increasingly important role in the fields of knowledge discovery and decision-making. Recently, researches and applications on learning Bayesian networks form data that have been developed are still new and evolving. Some organizations and magazines which focus on Bayesian research appear.The main achievements of this thesis include:1. This thesis makes a survey about the research on Bayesian networks, including the basic knowledge, the learning algorithms and the performance criterion of the learning algorithms, etc.2 Orienting edges of Bayesian network is part and parcel of learning Bayesian Network. An algorithm is proposed based on information theory and genetic algorithms. Cross-entropy is introduced on learning Bayesian network. Based on the network which oriented edges with cross-entropy, fitness function and genetic operators are designed, which provides guarantee of convergence. This algorithm can weaken the dependence of initial population and increase the convergence speed. Experimental results show that this algorithm can effectively orient edges of Bayesian network.3. Learning Bayesian networks structures based on Evolutionary Computing. There are two general approaches to Bayesian network learning from data, the dependency analysis methods and the search & scoring methods. Both of these approaches have their advantages and disadvantages. Although the search & scoring approach may not find the best structure due to its heuristic nature, it works with a wider range of probabilistic models than the dependency analysis approach. So now it is the most important task to devise a suitable scoring method and new evolutionary algorithms in order to improve the convergence and find the global optimum.Premature convergence is a serious issue in evolutionary algorithms since it might significantly degrade the overall performance. In this thesis, a brief analysis is given to the premature convergence phenomenon in evolutionary programming (EP) and two quantities which characterize the premature convergence are defined. Aiming at preventing and overcoming premature convergence, two approaches are presented. By combining the niche technology into the selection mechanism of EP and restart strategy into the framework of EP. we use niche technology to prevent the premature convergence phenomenon. while restart strategy is used to guide the research after premature. Then we learn the Bayesian network based on the two algorithms. To evaluate the performance of our algorithms, we conduct a series of experiments and compare them with previous work based on genetic algorithms (GA). The experimental results illustrate that both quality of the solutions and computational time of our algorithms are superior.4. Learning Bayesian networks structures based on particle swarm optimization. Particle swarm optimization (PSO), rooting from simulation of swarm of bird, is a new branch of Evolution Algorithms based on swarm intelligence. Concept of PSO, which can be described with only several lines of codes, is more easily understood and realized than some other optimization algorithms. PSO has been successfully applied in much engineering.Firstly; this thesis learns Bayesian network structure based on a binary PSO algorithms. Compared with GA; both quality of the solutions and computational time of the algorithm are superior. But PSO can be considered as the conventional particle swarm optimization; in which as time goes on; some particles become inactive quickly because they are similar to the global optimum and lost their velocities. In the following generations; they will have less contribution for their very low global and local search capability and this problem will induce the emergence of the prematurity. In this thesis; some concepts in immune systems are introduced into binary particle swarm optimization. Then we learn Bayesian networks based the immune binary PSO. Numerical experiments are compared with previous work based on GA. The results illustrate that the proposed algorithm not only improves the quality of the solutions; but also reduces the time cost.5. Applying Bayesian network in population problems. This thesis analyzes the factors which make impact on population quality from population data. Improving population science & culture quality decision-making model and improving population health quality decision-making model based on Bayesian networks are proposed. We analyze the probabilistic dependency relationship of the two models and give related results. In the end; the models and the results can be applied in population knowledge discovery and decision-making.