Research On Intelligent Medical Diagnosis System Based On Bayes Algorithm

With the high development of our society, each area starts to become intelligent.Various kind of intelligent systems appear in our life especially when date mining andmachine learning develop fast. The advent of this situation is caused by three reasons:First, it can help the experts to make diagnosis, and improve the accuracy of diagnosis.Second, it can simplify the flow of diagnosis, and improve the efficiency of diagnosis.Third, it can reduce the expense. The advent of intelligent systems make many areasbecome intelligence. Now many mature technologies have been applied in theintelligence systems. It has good social benefits.With the high development of standard of living, many diseases become verycommon. For instances, cardiovascular diseases, high blood pressure, etc. Thesediseases become the top killer of our life. The diagnosis process of these diseases is verycomplex. So it is very practical to develop an intelligence system to solve theseproblems. It is obvious that it can save time of diagnosis, reduce the expense andimprove the efficiency.In the area of data mining and machine learning, many mature technologies havebeen incorporated in intelligent systems. Such as decision tree, Bayesian, K nearestneighbors (KNN), k-means, etc. These methods can be divided into two steps: First,called training process. In this process, training data are used for construct modelaccording to the machine learning methods. After the model has been constructed, it canbe used for prediction and classification. For example, there are many diagnosis resultsmade by experts. This diagnosis information can be used for training. When a newinstance comes, the diagnosis can be made by machine automatically. These methodscan be classified into two catalogues: supervised learning, unsupervised learning. Ifeach instance has its own class label, and model is constructed on these data sets, thesemethods are called supervised learning methods. If all the instances have no class label,this kind of method is called unsupervised learning. That is to say, the instances belongto same class have maximal similarity. On the contrary, the instances belong to differentclass have maximal difference. In fact, there are still two ways of learning, lazy learning and eager learning. If the model is constructed each time according to each test instance,then this kind of method is called lazy learning method. If the model is constructed onetime, and each test instance is predicted by the same model, this kind of model is calledeager learning. Each kind of method has its own advantages in applications. Forexample, lazy learning methods are time-consuming but its classification accuracy ishigher than the eager learning methods. Eager learning is very fast in training process.This paper is focused on the application of Bayesian in medical diagnosis. NaiveBayesian is a lazy learning method. It has been used in many areas for its simplenessand efficiency. In fact it’s a classification method according to the probability ofbelonging to each class. This paper introduces the basic theory of Bayesian, and threekind of Bayesian: naive Bayesian, Bayesian network, optimum Bayesian. Generallyspeaking, the characteristics of Bayesian can be concluded as follows:(1) The basic theory of Bayesian is very simple, and it is very practical. If thepriori probability is known, the process of computation is very simple.(2) Bayesian is classification method based on probability. From this we canknow that, if the training data is not change large, then the classification results may notchange. That is to say, this method has good robustness.(3) The speed of learning is very fast.We design an intelligent system for medical diagnosis. As we know, there are twofactors which can affect the results: the number of attribute set, the number of cross.Different number of attribute set and number of cross may lead to different results. Sowe design different experiments to reflect the effect of different factors. There are twokinds of experiments: First is that, the number of attribute set is fixed, but the number ofcross is changeable. Second is that, the number of cross is fixed, but the number ofattribute set is changeable.We also develop the alerting system for patients. The basic theory is the differencebetween probabilities of belonging to each class. If the probability of belonging to goodis pretty bigger than the belonging to bad, then this patient may be very healthier. Onthe contrary, the patients may not be very healthier. In order to better describe the healthcondition of some person, we divide the health condition into four levels: excellent,good, middle, and poor. Under different conditions, some instances may belong todifferent levels. This kind of instances belongs to the edge of each class. Experiments results prove that this method is very practical. It has broader applications in medicaldiagnosis.