The Application Of The Mixed Kernel Function In The Fixed Term Deposit Of The Bank

One of the main businesses of a bank is the fixed term deposit. It plays an important role in the development of the bank. The data mining technology has been used in banking widely. As a newly and famous data mining method, SVM is applied into the deposit business to get a directive suggestion to the market.Support Vector Machine (SVM) has been developed rapidly as a new machine learning method. Kernel functions and the parameters of SVM have an important affection on the classifier. This study focuses on the combination of local, global kernels for SVM classifiers, applying global search for optimization of model parameters.The thesis consists of three parts:(1) A mixture of kernels is built by using a linear combination of the local kernel function and the global kernel function. This hybrid model of structure is an ensemble classifier.(2) AS a non-heuristic algorithm, grid search algorithm is used for the optimization of all parameters of the mixed-core model.(3) As a heuristic algorithm, dynamic particle swarm optimization algorithm is used for the optimization of all parameters of the mixed-core model.(4) For the reason of lower rate of ordering long-term deposits in the bank, the SVM model of the mixed core is used to get accurate customer division for the bank. Also with the regression model, we find the key factors impacting people’s decision, which gives directional proposals to the bank marketing.The innovation of this paper is as follows:(1) A SVM classifier combiner is made from the point view of mixing kernels;(2) It acts as a transformation from the single-core on the base of the grid search and MPSO algorithm, without increasing the complexity of the algorithm.(3) The speed of the MPSO algorithm is improved by the transformation on the formula of the accelerate convergence.