Application Research Of Data Mining Technology In Medical Diagnosis

In recent years,the number of diabetic patients in China has increased faster and faster,which has seriously affected national health.Studies have shown that sub-health is easy to induce diabetes,and diabetes is often accompanied by related complications.If you do not prevent or treat diabetes in time,it will bring a heavy burden to patients and the country.Sub-health lacks objective judgment rules and diabetes has a long incubation period,it is a big project to rely solely on manual diagnosis and brings great pressure to China’s medical system.Therefore,using data mining technology to train intelligent auxiliary diagnosis models has become a hot research topic.BP neural network and SVM are often used to establish disease prediction models.However,the BP network is easy to fall into the local extreme value and the SVM relies heavily on the parameters,which lead to unsatisfactory performance.Therefore,the thesis has taken BP network and SVM as the research objects,and established high-precision sub-health and diabetes diagnosis models.The main contents and innovations are as follows:(1)The basic principles of BP network and SVM have been introduced,and analyzed the existing improvement methods.(2)The diabetes dataset of the National Clinical Medical Science Data Center has been collected,and the data preprocessing work has been completed to generate training samples in a specific format.(3)Aiming at the premature problem of genetic algorithm(GA),an improved GA algorithm based on probabilistic perturbation and semi-initialization strategy has been proposed.The algorithm uses the perturbation strategy to increase the diversity of the population,and the semi-initialization fundamentally changes the limitations of the individual in the comparison process of the global optimal solution.Meanwhile,aiming at the local minimization problem of BP network,an optimized BP algorithm based on improved GA algorithm has been proposed and used to establish a sub-health state diagnosis model.Experiments show that the combined model based on improved GA algorithm and BP neural network has the best effect.The sub-health classification accuracy rate is 96.05%,which is 2.83 percentage points higher than the traditional model.(4)Aiming at the problem that Particle Swarm Optimization(PSO)is easy to fall into local extreme value and has severe oscillating in late stage,an improved particle swarm optimization algorithm(IPSO)based on dynamic weight and momentum term strategy has been proposed.The IPSO algorithm uses distance-based dynamic weights and Levi’s flight strategy to avoid premature convergence,and introduces momentum terms to slow down later oscillations.Meanwhile,aiming at the problem that SVM relies heavily on the parameter,an improved SVM based on IPSO algorithm has been proposed and used to train a diabetes diagnosis model.Experiments show that the proposed combination model based on IPSO algorithm and SVM has the best effect,and the classification accuracy rate for diabetes is 92.74%.(5)Based on the Java language and Spring MVC architecture,an intelligent auxiliary diagnosis system has been designed and implemented.In the system,users can log in and independently complete sub-health diagnosis and diabetes pre-judgment,which helps doctor making quick diagnosis.Meanwhile,the feasibility of the algorithm in practical application has been verified.