Detection of currency crises by a novel rule extraction method from support vector machine

Abstract

This study attempts to obtain a set of human comprehensible fuzzy if-then rules for the detection of currency crises from Support Vector Machine (SVM). SVM is used with explanatory variables known to be associated with currency crises to detect occurrences of currency crises. Fuzzy if-then rules are then obtained from the SVM through our novel rule extraction method which is called Support Vector Space Expansion (SVSE) method in order to unveil human comprehensible patterns behind SVM blackboxed system decision. The overall results of detection of currency crises of the fuzzy if-then rules are comparable to those from the SVM, and the if-then rules obtained may be used by financial experts to try to explain patterns of related financial statuses when currency crises occur, plus the if-then rules can also be easily incorporated into a software program using any popular computer language.