A Comparative Analysis Of Classifying Algorithms In Medical And Pharmaceutical Data

Classification problem is a wide range of problems in real life. In the field of medicine, medical diagnosis classification, adverse drug reaction monitoring, etc.It abound. Good classification algorithm can effectively solve the problems arising from the field of medicine, it can be better applied to decision support diagnosis, personalized medicine, precision medicine. Classification is based on the characteristics of data sets to construct a classifier, using the classifier of unknown categories of samples given category of a technology. The current classification algorithms varied, different classification algorithms will produce different classifiers, classifier adaptive direct impact on the efficiency and accuracy of the final classification results. Therefore, in the face of complex medical data, select the appropriate data classification algorithms in medicine is essential.Based on current pharmaceutically data classification algorithms comparative analysis is still rare. To compensate for the deficiency, the paper using breast cancer data, classification issues in-depth research, and model-based performance analysis and comparison of the characteristics of various algorithms. Comparative analysis on the actual UCI data found, SVM on multiple performance indicators can achieve good results, but its model complexity and higher, using a support vector 49, the use of high-dimensional space mapping associated with reduced relation. CART algorithm to construct a decision tree model by only three variables, to obtain a higher specificity, medical data reflects the characteristics of high correlation. KNN model will have the most excellent accuracy. Naive Bayesian model constructed by poor performance, mainly due to the independence of the dependent variable. But the overall results, a single classifier is difficult to have achieved good results in various performance indicators.Classification of medical or pharmacy is different from the other areas, clinically, the traditional binary classifier brings diagnostic error would have serious consequences, such as the false non-reaction conditions the reaction conditions, or the condition of non-reaction mistaken for the reaction conditions, and once such a misdiagnosis occurs, it would cause irreversible damage to the patient. So the traditional dichotomous classification application in the field of medicine has great limitations.Finally, based on the research of multiple classify methods, this paper puts forward a new kind of combined classifier(Fuzzy State-based Compound Classification Method, FuSBCCM). It is able to improve the accuracy of classification by abandoning or limiting the use of multiple fuzzy classification results inconsistent state brought about, It also learn the concept of entropy to measure and divide fuzzy state is conducive to further analysis or human intervention.Experiment performed on Wisconsin Breast Cancer shows that, FuSBCCM has better performance than any constructing classification method.