Research On Multi-Relational Decision Tree Algorithm

Multi-relational data mining is an important and rapid development of one of the areas of data mining in recent years. Efficiency and scalability of data mining have been an important research topic. Consider multi-relational data mining, the issue is particularly important. The complexity of multi-relational data mining put forward to higher requirements on performance of the algorithm. With the traditional data mining algorithm, the search space of multi-relational data mining algorithm becomes more complex and much bigger. For multi-relational data learning algorithm, the main bottleneck of improving algorithm efficiency is the assumption space. In view of the above problems, this paper do the following work:First of all, this paper has studied on data mining theory, multi-relational data mining theory , Especially it has studied deeply on multi-relational classification algorithms,multi-relational decision tree and latest technology of multi-relational data mining -tuple ID propogation.Secondly, this paper has proposed an improved algorithm based on multi-relational decision tree. The improved algorithm based on multi-relational decision tree has been improved in two major areas: 1 In order to improve scalability of multi-relational decision tree algorithm, tuple ID propogation technologies of virtual connecting has been applied to improved algorithm based on multi-relational decision tree;2 In order to reduce the time of searching by system alone,reduce time of the system searching useful attribute alone and increase user's satisfaction,it has proposed a technology of the delivery of background attributes of completing the task of classification under the user's guidance.this technology has been applied to improved algorithm based on multi-relational decision tree. Finally, In this paper, it has gave the theoretical proof and the experiment proving of the improved algorithm based on multi-relational decision tree. This paper has used three relations (Loan, Account, Transaction) of PKDD CUP'99, and has done compareing experiment with two methods between improved algorithm based on multi-relational decision tree and multi-relational decision tree.The first methods is that the number of records in the three relations remain unchanged,it did the experiment with the number of property increasing in every relation; The second methods is that the number of property in the three relations remain unchanged,it did the experiment with the number of records increasing in every relation.Through the above experimental results, This paper consider that,when the improved algorithm based on multi-relational decision tree does not meet the threshold of the background of property transfer in the search data item, the improved algorithm based on multi-relational decision tree has lower operating efficiency; when the improved algorithm based on multi-relational decision tree meets the threshold of the background of property transfer in the search data item, the improved algorithm based on multi-relational decision tree has relatively higher operating efficiency and is less affected by increasing the number of property (or increasing in the number of record).