Improving Text Mining Methods in Market Prediction via Prototype Selection Algorithms

Abstract

<strong>Nowadays, researches are faced with large volumes of data. Since </strong><strong>a considerable </strong><strong>amount of them are unstructured, they cannot be processed naturally. </strong><strong>Hence</strong><strong> two main challenges in this field are high dimensional of feature</strong><strong>s</strong><strong> space and </strong><strong>bulk </strong><strong>of available data. In this research, a feature selection method based on target features </strong><strong>is propose</strong><strong> to handle the curse of dimensionality. Moreover, </strong><strong>to address </strong><strong>the huge volume of data some of prototype selection approaches are utilized. The proposed method in this paper has three essential steps that each step improves the previous ones. Although, the proposed method </strong><strong>reached </strong><strong>significant results in each phase separately, its best performance obtained via the last phase in terms of classification accuracy rate. To evaluate the performance of the proposed method, it has been compared with three-layer algorithm. The results </strong><strong>revealed</strong><strong> that the proposed method had </strong><strong>significantly</strong><strong> better results than the three-layer algorithm in average</strong><strong>.</strong> <br />