The Study Based On Data Mining Of Financial Risk Early Warning Classification

The study in pre-warning system of financial risk occupies a significant position in the field of financial data-mining. Due to its diversity, complication in interrelationships, innate dynamics and hugeness in amount, as well as its noisiness and non-normality, which are general characteristics of financial data. The study of financial early-warning appears to be especially challenging. Implementing the technique of data-mining to discover the underlying laws within the innumerable financial data can efficiently reduce the operational risks of financial institutions. Therefore, the study in pre-warning system of financial risk has an extensive application value and market prospect.This thesis is divided into six parts, which is organized as follows:Chapter one would firstly offer some background information of the research topic, as well as its implications; and based on the review of the present studying situation in this topic home and aboard, this part will bring out the aim and implication of this thesis.Chapter two mainly proffers fundamental conceptions relating to financial risk and the general procedures of early-warning system.Chapter three analyzes the feasibility of implementing data-mining in the field of finance; by making a comparison between traditional mode of data-processing in financial area and data-processing applying the technique of data-mining, and analyzing the peculiarities of financial data, the value of implementing data-mining in the financial area thus becomes obvious.Chapter four focuses on the study of fuzzy clustering model; while adding clustering number and index into traditional mode, this part suggests ways for improvement. Slow converging speed, the existence of a local minimum point in target function, the difficulty in determining the hidden layer number sand hiding nodes are all shortcomings of BP neural network models. While taking these into consideration, this thesis proposes to employ the improved BP model attached with the momentum and to implement a hybrid of those two models in the field of early-warning in financial risks.Chapter five, according to Chapter four of the hybrid model to establish an empirical analysis, and with the algorithm analysis comparing the results of the BP neural network the results show the superiority of the hybrid model.Chapter six is the last chapter of this thesis, which summarizes the whole thesis; meanwhile, it points out the existing problems as well as its research direction in future.