Study of temporal data mining technique on Dhaka Stock Exchange

Abstract

One of the major tasks in stock market analysis is the discovery of specific events that give rise to a particular event. In this research, we emphasise on temporal data mining conducted on a stock-oriented dataset with a time dimensional approach. We are mainly interested in bringing forward an algorithm for pattern discovery in sequential data streams and also bring out the interdependencies among the events. This led us to the discovery of sequential continuous patterns. The patterns serve as rules that enable us to determine the occurrence of an event on a particular stock-transaction day. In our paper, we have proposed and implemented the STRDTM (stock trading rule discovery by temporal mining) algorithm with real life data from Dhaka Stock Exchange (DSE) as input. We generate high confidence rules from training dataset of six months and finally evaluate the performance of the discovered rule on testing dataset. On an average, we could achieve around 80% accuracy if we trade on the discovered rules.