Study Of Data Mining And Its Application In Financial Early Warning

Since the last few years of the 20th century, rapid expansion in global capital markets and the growing popularity of online trading make it crucial to investors and managers to forecast the company's financial crisis as early as possible. For the former, it gives investment opportunities and avoids investment risks; For the latter, it provides information for decision-making and market-occupation. Therefore, financial early warning is becoming more and more significant in the fields of management and economics.Followed the rapid development of information technology and other related research fields, data mining has been an important financial early warning tool, and many study results also show that the forecasting models based on data mining are superior to the forecasting models based on traditional statistics. However, the data mining technology is still in the developing stage. This dissertation combines new ideas to data mining and applies it to financial early warning. Uncertainty theory has been a focus to scholars in various fields including data mining. In this dissertation, fuzzy set and rough set have been utilized to improve traditional data mining, and the new data mining methods are used to set up forecasting models on financial early warning. The dissertation also makes use of the data of Chinese listed companies to prove the effectiveness of the models. The main ideas of the dissertation include three aspects:Firstly, the dissertation analyzes the limitations of the traditional neural networks which are extensively used in the financial early warning. A fuzzy neural network model based on attribute reduction in rough set is put forward to overcome the limitations. The model makes use of rough algorithm in data preprocessing and fuzzy neural network in forecasting. The training data and the testing data of the model are the Chinese listed companies which fall in financial crisis in 2005 and 2006. The result shows the effectiveness of the model.Secondly, two clustering methods based on the uncertainty theory are put forward. One is the fuzzy clustering algorithm based on particle swarm optimization. The other is a K-means clustering algorithm based on rough sets. The traditional clustering algorithm seldom been used in financial early warning, but the dissertation analyzes the reason and gives a solution on it. The disserta- tion also establishes two new financial early warning models by means of the two clustering algorithms.Lastly, the dissertation analyzes the support vector machine in financial early warning, and points out that special knowledge is useful to forecast the financial crisis. A fuzzy support vector machine which could incorporate expert's knowledge is introduced. Qualitative and quantitative analysis will be combined to form a new forecasting model. The experiment shows the model can improve prediction capabilities of the financial early warning. In the end of the dissertation, the characteristics of the three financial early warning methods are summarized.