Recognition of pivotal instances from uneven set boundary during classification

Abstract

Database may contain pivotal records-small chunks of records or instances consist of important information specific to the domain. These chunks of instances may contain crucial information which assists in decision making by assigning labels to pivotal records, unlabeled data instances and improves accuracy of the classification model. Our work suggests the heuristic Rough Set Boundary detection for approximating the boundary set efficiently from the large database to reduce the search space substantially for finding critical records. The use of Rough Set Boundary detection has the advantage of obtaining rough set from the original data set which confines the search space only to the boundary. It uses the concept of pivotal score for each instance in the boundary to isolate the critical records. The method also exploits Feature Selection technique for reduced set of attributes in order to obtain less computational time. The proposed work retrieves the pivotal records from the boundary set and also improves the classification accuracy by increasing true positive and true negative errors. Experiments are carried out for real—world medical data sets with numeric values and various classification algorithms are executed to validate the results. Result shows that the identification of pivotal records from rough boundary set helps for improved classification accuracy using less computational time and which are validated using real-world data sets.