A New Approach for Exceptional Phenomena Knowledge Detection and Analysis by Data Mining

Learning logic of exceptions is a considerable challenge in data mining and knowledge discovery. Exceptions are the rare phenomenon with positive unusual behavior in a database. Creating an efficient framework to increase the reliability in the detection of exceptions in the knowledge and learning is quite important. This paper presents a novel framework to promote the confidence to a limited number of records (exceptions) for effective learning of exceptions. In this study, a new approach based on the abnormality theory and computing theory is presented to detect exceptional phenomena and learn their behavior. First, Renyi entropy function is implemented to detect exceptional data which is differentiated data according to their hidden knowledge. Then, the novel E-RISE algorithm which follows bottom-up learning strategy is introduced to learn exceptional data behavior. Efficiency of the proposed model is determined by applying it to the Iran stock market data. Mining the number of 1334 stocks data points, 2.6% of them had exceptional behavior. The extracted rules represent the exceptional stocks attitudes. After that, an expert system is designed to use the extracted knowledge for recognizing new exceptional stocks. Faced with new stock, this expert system can recognize exceptions by comparing its characteristics with normal and exceptional behavior. Exceptions behave in compliance with exceptional rules or in contradiction with any normal pattern. This acquisition knowledge is the basis of exceptional portfolio selection which aims to make exceptional wealth for investors. Findings of the proposed method are compared with the outcomes of applying traditional methods as decision tree and support vector machine which is considerable. The results show the capability of the proposed method in exceptional data detection and learning their behaviors.