26bresearch On Relevance Vector Machine And Its Applications In Data Mining

In the field of artificial intelligence and data mining, machine learning method is of significance to research. Relevance vector machine(RVM) is a novel machine learning method to classify and predict. It has more advantages than other algorithm, such as high sparsity, strong generalization ability and kernel function need not meet the Mercer condition, etc. However, in the research of the existed RVM model, it is the problem that the kernel functions can not be chosen scientifically and rationally, which result in low classification prediction accuracy and efficiency. To solve the problem, multi-kernel relevance vector machine method based on multi-kernel functions is mainly researched, and it is applied into the oil well logging data mining.The main work or innovations are as follows.(1) The analysis and simulation on the multi-kernel RVM method based on evolutionary computation: the basic principles of classic RVM is analyzed, multi-kernel RVM model is built in accordance with multi-kernel idea,and the multi-kernel RVM method based on evolutionary computation is researched, ie. particle swarm optimization(PSO) is adopted to optimize the parameters of multi-kernel RVM method. Typical simulation results show that the classification accuracy of the multi-kernel RVM based on evolutionary algorithms is higher than that of classic RVM and PSO-based RVM, and the PSO-MK RVM method can find out the optimal kernel parameters and combinations of kernel function, but there are some shortcomings needed to improve, such as long training time, low stability and so on.(2) Research on multi-kernel RVM based on Second-Order Cone Programming(SOCP-RVM): To solve the problems in classic RVM and PSO-MK RVM, such as the low classification accuracy, much operation time and poor stability, a novel multi-kernel Relevance Vector Machine method based on SOCP algorithm(SOCP-RVM) is presented, ie. multi-kernel RVM model of the multi-kernel alignment based on SOCP algorithm is built,and it is inferred that multi-kernel RVM model transformed into a convex optimization problem solved by SOCP algorithm in theory. The simulation results show that the presented method can improve the classification accuracy and greatly cut down the operation time.(3) Application research on oil logging data mining. To improve the effect of oil well logging data mining, SOCP-RVM oil layer recognition model is built to identify oil layer of the oil well in Xinjiang. The achieved results show that the effectiveness and operational efficiency of the SOCP-RVM method are superior to the classic RVM and improved methods, and the data mining results are remarkable in the oil layer recognition.