RS-S3VM: Combining Rough Sets with S3VM for Image Classification

In the landscape of data mining applications, the difficult task of manually labeling extensive datasets presents formidable challenges owing to its inherent difficulties, costliness, and time consuming nature. To address the limitations associated with manual labeling, semi-supervised learning emerges as a promising paradigm, harnessing the potential of both Data with and without labels in the course of training. SVM based semi-supervised learning (S3VM), one of the several semi-supervised learning approaches, has shown potential in improving performance of classification. This research paper introduces a novel approach that combine rough sets with SVM based Semi-supervised learning for image classification. This innovative methodology involves employing a Convolutional Neural Network (CNN) model to undergo training on labeled data, make predictions on unlabeled data, and subsequently undergo retraining. Feature extraction is then executed for classification. Following this, Rough Set theory based feature selection technique is used, generating a reduct that functions as a rule for SVM. This comprehensive method not only improves the field of image classification but also offering a novel and pragmatic solution to problems in semi-supervised learning. This procedure concludes in an evaluation of the models final out come using experiments on the CIFAR-10 benchmark dataset, where it achieved an impressive accuracy of 97.30%.