The Application Of Bayesian Network Based On Tabu Search Algorithm In Diseases Prediction And Diagnosis

Objective:To study a new method called tabu search algorithm for global optimization, and explore the effect of the global optimization algorithm to bayesian network. Then selecting a proper method to build the bayesian network model of clinical diseases, which can discover the influence factors of the diseases and predict the probability of clinical diseases for diagnosing disease differentially, so as to we can reduce the occurrence of misdiagnosis or missed diagnosis and provide scientific and reasonable evidences to disease prediction and diagnosis. Methods:First we use three methods of Hill-climbing, K2 algorithm and tabu search algorithm to construct the bayesian network model respectively, and compare the advantages and disadvantages of different models from the aspects of run time and evaluation score. Then we compare the bayesian network model with the traditional logistic regression model about effect factors of coronary heart disease to evaluate the effect of different models, and predict the probability of disease. Finally we utilize the bayesian network model by the acute inflammation data that downloaded from the University of Californian Irvine to diagnose acute inflammations of urinary bladder and acute nephritises. Results:(1) Generating different size simulation data of the Chest’s Clinic model by utilizing the Netica software, and using Hill-climbing, K2 algorithm and tabu search algorithm to study the bayesian network respectively. The results suggest that the larger the data size, the more accurate the bayesian network structure determined. But the bayesian network built by tabu search algorithm is better than Hill-Climbing and K2 algorithm in different sample sizes, and the bayesian network structure built by the tabu search algorithm is almost the same as the original network when the sample size is 1000;(2) Using tabu search algorithm to construct the bayesian network model for analyzing the risk factor of disease, and comparing them with the results of logistic regression model analysis. It shows that the bayesian network constructed by the tabu search algorithm can objectively describe the relationship between effect factors and disease, and it can represent the direct and indirect relations of variables through the network. So the network expresses the quantitative and qualitative data information at the same time, and makes the results more visual and intuitional;(3) Using the bayesian network constructed by the tabu search algorithm to analyze the acute inflammation data sets from the University of Californian Irvine, we found that the bayesian network constructed by the tabu search algorithm is consistent with the results from the relevant literatures, and it is useful to diagnose diseases accurately. Conclusions:The results in this paper show that the optimization effects of the tabu search algorithm to bayesian network model is good and reliable. The bayesian network is helpful to determine the internal relationship between things, and found the developmental rule of things. It can be applied in clinical practice, and has important application value in disease prediction and diagnosis.