Using feature selection and classification approaches in credit score and spam filtering

• Main Authors: Yen-Wei Su, 蘇彥暐
• Other Authors: Ping-Feng Pai
• Format: Others
• Language: zh-TW
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/27394857735362845888

碩士 === 國立暨南國際大學 === 資訊管理學系 === 95 === Data mining is a popular technology recently. For example, it can apply to spam filtering, credit risks, medical, financial forecasting and industry, etc. The technology can forecast various situations precisely and effectively. The classification technique is a important domain in the data mining. In our research, we used these tools to filter spam and the credit card, and combined the feature selection technology to find the important attributes from original data to improve the accuracy. Credit score is an index to judge the risk of break with applicants, the purpose avoids the loss of company or banks. But from the large number of customer’s database, it may cause the error from the noise (irrelative attribute). Hence, the experiment brings up a hybrid method, using factor analysis to filter the noise before classifying in my experiment. Besides, a major research is spam filtering. Many filtering techniques such as Back propagation network, Decision tree, Bayesian filtering or Support vector machine etc., resist the spam because of spam always besets users. And the filtering targets often different, likes as black white list, heading of mail or contents of keywords in the mail. So, we hope to combine the feature selection and filtering methods to improve the accuracy of classification. The content of the mail is the target in this experiment, it including: 1. The keywords in the mail, 2. The heuristic feature and 3. Both. The research will compare the three conditions and different filtering approaches, and get the best result of classification. The two experiment results indicate that using feature selection and classification approaches can improve the accuracy and reduce the noise.