Decision Tree Classification Algorithm Based On The Correlation Function

Data Mining, which is also called knowledge discovery in database, is a process which is used to automatically refine useful patterns from data set. Classification and prediction is one of the main subjects of data mining, while decision tree algorithm is the most popular classification algorithm among all classification algorithms.But most of the existent decision tree algorithms suffer a problem, namely multivalue bios, in the process of attribute selection. Multivalue bios may result in inducing wrong knowledge from data set, and consequently result in the decline of the performance of decision tree. According to the problem above, this paper mainly illustrates three problems. First, this paper analyses the multivalue bios problem of decision tree algorithm experimentally and theoretically. Second, this paper proposes a new decision tree algorithm, AF algorithm, which avoids multivalue bios. Third, we implement the AF algorithm, and use it to classify patients in the field of iatrology.What is new in this paper? First, this paper proposes a theoretical method for analysing the multivale bios problem of decision tree algorithm. Experiment is the traditional method for analysing multivalue bios of decision tree algorithm, but it has a fault that we must have the expertise of the specific field. This paper proposes a theoretical method, with which we can analyse the multivalue bios problem without the expertise of specific field. Second, this paper proposes a new decision tree algorithm, AF algorithm, which is based on association function. On the basis of analysing the multivalue bios, this paper proposes a new decision tree algorithm, AF algorithm, which avoids the multivalue bios problem. Through evaluation, we find AF algothm is better in many aspects than ID3 algorithm which is widely used in many fields.