Text Categorization Using a Novel Feature Selection Technique Combined with ELM

The rapid growth of digital documents and internet users on the web increases the searching time of a document for the end user, which affects the performance of the search engine badly. Hence, to reduce the searching time and to increase the efficiency of the search engine, text classification is the need of the day. But to do an efficient text classification, selection of good features also equally important. To address this issue, the current paper proposes an approach called Combined Correlation Discriminative Power Measure (CCDPM) where first the highly correlated terms (features) are removed from the corpus and then using the scores generated by discriminative power measure technique, the uncorrelated features of the corpus are ranked. Top k features are selected to generate the reduced training feature vector. For classification, Extreme Learning Machine (ELM) is used, and the empirical results on four benchmark datasets show the efficiency the proposed approach compared to other state-of-the-art feature selection techniques. Results of ELM are more promising compared to other conventional classifiers.