Research On Self-Assistance Registration System Based On Bayesian Incremental Learning Model

With the development of CAD due to the introduction in medical care sector of the techniques like data mining, machine learning and pattern recognition, the prospective application of CAD is increasingly encouraging more research on bio-medical machine learning. Also, the development of software and hardware has boosted this research. The techniques such as intelligent medical diagnosis, medical imaging, machine vision can exemplify this trend. Among these applications, medical diagnosis has become the most important area benefiting from data mining and machine learning. Many classification techniques such as decision tree, the nearest neighbor and kernel method have gained popularity. This paper, guided by machine learning theory, analyses deeply the data and condition of the patients, refers to, modifies and improves the classical model and algorithm, and then applies them in the self-assistance diagnosis system to meet the patients’ demands, thus improving the medical diagnosis quality and efficiency.This paper conducts a deep research on methods of medical intelligent diagnosis, and explores with a special attention the Bayesian classification techniques, the model on-line learning, and minimum decision risk theory. Then, this paper improves Bayesian incremental model and the corresponding algorithm, and based on this model, designs a self-assistance system with the function of intelligent diagnosis. This research has innovated the following fields:1) With an understanding of the patient’s data as its basis, and the classical machine learning method, this paper proposes three improvements in Bayesian incremental model. First, this paper describes the incremental leaning method of instance with new label and new attributes. Second, instance selection strategy is introduced to train the model. Last but not least, the minimum risk decision is introduced in the predicative method. Incremental learning algorithm can establish a initial model by the previous data and, with the increasing data, can select by itself the instances for the improvement of the model prediction and modifies the data accordingly, and then a new model comes into being. Compared with the Maximum posterior probability method in Bayesian classification, the minimum risk decision can balance the increasing data in their application. This paper tests the data on the UCI data section, and evaluates the predictive effects with accuracy and recall rates as two indexes. The test illustrates that the modified model can gradually improve the classification function of the system and can adapt to the new label and new attributes.2) This paper, based on the improved incremental Bayesian model, designs a self-assistance registration system with a function of recommendation of hospital departments. This system first gains predictive model from the patients’data and, with this model, justifies the departments that a patient is supposed to turn to and as the newly added data modify the model, the patient’s anxiety and the hospital cost can be saved. In addition, the increasing data can contribute to the better understanding of the patients.