The internet is teeming with an ever-increasing amount of text information, which can come in various forms such as words, phrases, terms, patterns, concepts, sentences, paragraphs, and documents. The vast quantity of data can pose a difficulty in terms of organizing and structuring textual data effectively. In existing research work, imbalance in counting the terms hampers the classification results. We prioritize the data that precisely fits into the correct class to reduce the imbalances in the dataset and improve the overall result quality. Significant improvements are noticed in accurately classifying text by maintaining an adequate ratio of text data and using efficient text classification approaches. To improve the generalized ability of ELM, feature Selection and optimization of Deep Learning algorithms produced a great influence on classification. In this paper, the Enhanced Relative Discriminative Criterion (ERDC) and Ringed Seal Search along with Extreme Learning Machine (RSS-ELM) have been proposed for text classification. Experiments are conducted on three text datasets named: Reuter21578, 20 newsgroups, and TDT2 with a different number of classes, which shows proposed ERDC technique presents an average of 91.6% accuracy results among the previous IRDC & RDC techniques. Moreover, the proposed RSS-ELM produced a significant result of around 99.1% as compared to existing CS-ELM and GA-ELM techniques which count an average of 66%, and 54% respectively.