Mining linguistic knowledge from financial data

Abstract

This thesis proffers a cloud-based linguistic approach to discovering qualitative patterns and linguistic rules from continuous data, with a special emphasis on its applicability to financial data analysis. The linguistic strategy is achieved by the cloud technique, coupled with hybrid linguistic data mining algorithms. The cloud technique constitutes the corner stone of the linguistic approach, which serves as an operational bridge between quantitative and qualitative representations. With this technique, cloud model is used to represent continuous data in terms of linguistic variables that are defined by three numerical parameters, taking full account of the fuzziness and randomness of the original data. Meanwhile, a cloud-based data clustering algorithm is developed to auto-generate proper linguistic variables from continuous data, through an iterative optimizing process drawing on the Genetic Algorithms. In order to discover different kinds of linguistic knowledge, a hybrid strategy is employed by integrating the cloud technique into various established data mining algorithms. Three hybrid linguistic data mining algorithms, namely, cloud-based attribute-oriented generalization algorithm, cloud-based C4.5 decision tree algorithm and cloud-based Apriori association-rule algorithm, are developed in the thesis. These hybrid linguistic data mining algorithms can effectively handle continuous data in the mining process, and represent discovered results with domain-oriented and user-understandable linguistic concepts. Moreover, a validity preservation technique is proposed as a supplement to the linguistic solution, which detects and refreshes outdated linguistic variables and invalid linguistic patterns, so as to secure the validity of discovered knowledge in a rapidly changing data environment. In addition to its theoretical and methodological novelty, this thesis also adds value to the application profile of data mining. It exhaustively explores the applicability of linguistic data mining techniques to financial analysis, taking Hong Kong stock market as a case in point to illustrate its potential for business applications. Such empirical studies evidently demonstrate the many merits of linguistic approach: excellent understandability, outstanding capability to deal with large continuous data sets, and high tolerance and robustness in noisy and changing data environment. It is suggested that linguistic strategy could be a promising solution to a number of challenges faced by data mining researches.