Knowledge acquisition from an incomplete domain theory ? An application on the Stock Market

Abstract

This paper describes an application of a machine learning technique, the Induction Over Unexplained (IOU) approach, which combines Explanation Based Learning (EBL) and Similarity Based Learning (SBL) methods to acquire concept descriptions from a set of training examples. IOU is designed for efficient learning with incomplete domain theory and partially explainable features of training examples. In this paper, the IOU approach is used to generate rules that can later on predict the future performance of stocks. There are many factors that can affect the price of stocks. Some of them have a direct relationship with the change of a company's stock price, and can be easily explained and modelled using the EBL method, while others may have only an indirect relationship for which the SBL method is preferable. In other words, a good rule which can accurately predict the future performance of stocks should include both explainable and unexplainable features. Finally, the performance of the learning results of the IOU approach was compared with that of a similarity based learning system and two other statistical methods. The results showed that the IOU method outperforms all the other approaches especially when the number of training examples is small.