Analysis Of Observations Daily Living Based On Bayesian Network

The daily lives of people become increasingly diversified nowadays,their health conditions and living styles are influenced by daily activities,such as sports,online-shopping,games,reading,social and other daily behaviors.With the development and popularization of the Internet and intelligent devices,it is convenient to collect the data generated by daily activities.We can discovercertain personal information hidden in the data,and the information will effectively improve the quality of people's lives.As a widely used uncertain knowledge representation model,Bayesian network provides a way to describe the causal relationship among variables.In this thesis,two key technologies,Bayesian network structure learning algorithm and joint tree algorithm,are studied and applied to analyze the observations of personal daily lives.Considering the disadvantagesof the existing Bayesian network structure learning algorithms,two improved algorithms are proposed,the hybrid structure learning algorithm and the incremental learning algorithm respectively.Most of the existing structure learning algorithms do not consider graph connectivity completely and have some undirected edges,which leads to low accuracy of learning results.In this thesis,a hybrid search algorithm(NITDO)based on greedy search strategy is proposed,which combines the dependency analysis method and the score-search method.The new method improved the accuracy of learning resultssignificantly.Since the amount of data about daily lives increase as time goes on,an incremental structure learning algorithm(ISL)is proposed based on the idea of score-search algorithm.By fully using of the structural information that have been learned,the accuracy and efficiency of the algorithm are further improved.Meanwhile,based on the incremental learning algorithm,a multi-stage learning method for static data is designed and implemented.The junction tree algorithm is an accurate inference algorithm for Bayesian network.And the triangulation operation is one of the key steps,which is related to the deletion order of nodes.As the simple genetic algorithm' sconvergence rate is relatively low and its results could easily fall into be premature,an adaptive genetic algorithm(TAGA)is proposed.Algorithm TAGA selects individuals by using improved linear sorting selection method.And by the use of a new adaptive crossover operator and a population migration strategy,the performance of convergence is improved,and the premature problem is alsorelievedeffectively.Finally,for the purpose of illustrating the effectiveness of Bayesian network in processing daily living,we give an application example of the daily life.