High Accuracy Data Classification and Feature Selection for Incomplete Information Systems Using Extended Limited Tolerance Relation and Conditional Entropy Approach

Data classification and feature/attribute selection approaches play important role in enabling organizations to extract meaningful insights from vast and complex datasets. Besides, the accuracy and processing time are two parameters of interest to determine which approach is favourable or suitable for enormous data. Moreover, the presence of redundant, incomplete, noisy and inconsistent data made more concern to accuracy and computational resources. The issue of incomplete data is addressed in limited studies due to its complexities, particularly on data classification and accuracy as well as attribute selection. The limited tolerance relation between objects is the favourable approach used in this scenario. However, the accuracy and the data classification rate need to be improved. In this paper, a new approach called extended limited tolerance relation with the similarity precision among objects to improve the data classification with high accuracy will be presented and the feature/attribute selection is performed using conditional entropy. Comparative analysis and experiment result between the proposed approach with limited tolerance relation approach in terms of data classification and accuracy are presented. The proposed approach comparatively improved the accuracy with better data classification rate and feature selection while preserving the consistency of the information in incomplete information systems that is worthy of attention.