A Fuzzy Feature-Selection and OBL Chef Optimization Framework for Cancer Image Classification
Contributors
Bibhu
Subrat
Keywords
Proceeding
Track
General Track
License
Copyright (c) 2026 Sustainable Global Societies Initiative

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Abstract
Because dermoscopic pictures show significant visual variability, class imbalance, noise, and overlapping patterns between benign and malignant lesions, reliable skin cancer image categorization is still difficult. This paper introduces CAF-FmRMR-OBL-ChefNet, a hybrid framework that integrates opposition-based learning Chef Optimization Algorithm (OBL-COA) for feature-subset optimization, fuzzy minimum redundancy maximum relevance (FmRMR) feature ranking, multi-architecture deep feature extraction, and image preprocessing. The presentation details a dataset of 10,605 photos of melanoma skin cancer, 9,605 of which are utilized for training and 1,000 for testing; the images are cropped or shrunk to 224 × 224 pixels in RGB format. While complimentary deep representations are extracted using ResNet50, ResNeXt, and NASNet, images are enhanced using Gaussian filtering. The fused features are ranked by fuzzy relevance-redundancy scoring, and OBL-COA looks for a compact subset that is assessed using classifiers like SVM, AdaBoost, Gradient Boosting, and Extreme Learning Machine (ELM). Mean accuracy of 0.9016 ± 0.0046, balanced accuracy of 0.9013 ± 0.0047, ROC-AUC of 0.9586 ± 0.0027, and an average of 29.9 selected features, which corresponds to a claimed feature reduction of 70.1%, are among the reported summary statistics for the suggested experimental framework. According to the presentation, SVM has an accuracy of roughly 90.2%, making it the top classifier overall. These findings imply that feature dimensionality can be decreased without sacrificing effective classification performance by combining opposition-based optimization with fuzzy feature ranking.