Mining financial time series databases using machine intelligence and evolutionary computing techniques

Abstract

This thesis has taken up financial time series forecasting as one of the key areas of data mining tasks. Financial time series is understood to be volatile, random and chaotic. Three vital areas stock market, currency exchange rate and electricity prices have been studied in this thesis. Interesting to note, in this globalized economy, they assume an important position in the realm of international finance and domestic economic policy. They shape and are shaped by them. Extraneous factors as varied as political upheaval, terrorist attack or weather conditions also contribute to the movement of these systems. This dynamism and complexity makes their study intellectually challenging; hence, it is interesting to know why they behave as they behave. To this end in view, this thesis has sought to use machine intelligence and evolutionary computing techniques. It has experimented with various tools and techniques; hybridized them; compared them; to find out a suitable solution that can forecast the possible future movement of these dynamic and complex time series. Specifically, a PSO based Functional Link Interval Type-2 Fuzzy Neural System (FLIT2FNS) model for stock market; a hybrid Evolutionary Functional Link and Interval Type-2 Fuzzy Neural System (EFLIT2FNS) for currency exchange rate; and a Dynamic Filter Weight Neural Network (DFWNN) using a sliding mode weight adaptation technique integrated with DE for electricity prices forecasting have been applied and are found to have been good solutions in terms of producing accurate result. It can reasonably be said that these techniques can prove useful, at the least, for short-term forecasting of real- life financial time series.