Data Mining Solutions to Finance
|
Data Mining Solutions to Finance |
Objectives:
There are a number of objectives in mining financial data, including
 |
|
|
build mathematical models that can be used to predict market prices |
|
quickly identify clusters from data, such as survey data |
|
trend forecast (linear, seasonal, uncertain and random components) |
|
demand forecast based on customer satisfaction (consumption utility) |
|
Consumer behavior analysis (satisfaction, preference, subjective appreciation, etc.) |
|
correction and association analysis to relate targets to factors |
|
Feature Selection:
How to select and use financial factors to describe the underlying operation of a financial system is a complicated problem. People use the theory of micro- and macro-economics, econometrics, to select features to describe financial systems.
Up to now, features in financial analysis are extracted by human intelligence. Financial system structure is very complicated and it is often described by a large number of factors. The difficulty is how to choose the right set, or a reduced set, of the factors that correlate the financial activity with the structure of financial systems for maximum return. It seems that such empirical rules can also be discovered by the MasterMiner™ software that has implemented a few powerful methods for feature selection and feature reduction purpose.
The separability criteria, implemented in Zaptron’s MasterMiner software, are rather useful in selecting key factors that influence the performance of business. People often use linear and nonlinear regression in data separation with poor results. Zaptron’s data mining software is useful in simplifying the selection of nonlinear terms in regression. It has been compared favorably against the popular JMP and RS-1 software. The former uses far less terms in mathematical models, and produces lower prediction error (PRESS).
Traditional Methods:
Traditional methods include those from classical statistics, including linear and nonlinear regresion analysis, factor analysis, correlationa and association, and those from artificial intelligence, including neural networks, fuzzy logic, genetic algorithms, and so on. In pattern recognition applications, PLS (partial least square) method is usually used to find quantitative target-factor relationships. However non-linearity exists among target and factors which calls for new methodology.
Zaptron’s MasterMiner™ software has a number of major advantages over those based on pure ANN or pure regression (such as PCA) that are currently used in financial data mining. In addition, the separability test & classification capability by MasterMiner™ has also been proved to be very effective in practice. Examples include
|
|Home|Products|Application Notes|Research|Contact|
back to Home, Email to info@zaptron.com.
© 1998-2000 ZAPTRON Systems, Inc.