A Study of Using Random Forests Algorithms for Filtering Candidate Variables on Building a Multinomial Logistic Regression Model of Investors’ Risk Tolerance- A Case Study of Bank A

• Main Authors: Wen-Lin Yang, 楊玟霖
• Other Authors: Te-Hsin Laing
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/sr838v

碩士 === 輔仁大學 === 統計資訊學系應用統計碩士班 === 103 === Due to the popularity of Internet and mobile communications, Global information technology is an inevitable trend. In recent years, We had greatly advanced on field of data mining and machine learning., so many algorithms can achieve highly accuracy. Hence, their application is quite extensive. In machine learning, random forests have many advantages, such as, deal with thousands of explanatory variables, produce highly accurate classification, multicollinearity is insensitive, be able to build the strong classifier for unbalanced data or missing data, assess the importance of variables. And therefore random forests is the one of the best algorithms. Random forests contain more than one decision tree, each tree represents a tree structure and properties, the objects be classified via its branches according to their type. Random Forest is the classification of these trees, select the highest repetition of degree as a results of decision variable. The study try to use random forest to select the stabilize variable, and then substitute the stabilize variable into a Multinomial Logistic Regression, hope this method can produce the most streamlined variable on building a model to get better prediction than substitute whole variable into Multinomial Logistic Regression directly. To achieve authentication the above , the study want to build a model of Investors’ Risk Tolerance by customer information from domestic financial holding bank of wealth management ,and questionnaire of investment behavior.