Lightweight Hybrid CNN-WOA-ML Model for Automated Colon Disease Diagnosis

This study proposes a novel hybrid approach for colorectal disease detection from histopathological images. The novelty lies in integrating lightweight convolutional neural networks (CNNs) (such as SqueezeNet) with Whale Optimization Algorithm (WOA) for feature selection and Support Vector Machine (SVM) for classification, achieving high accuracy with low computational complexity. SqueezeNet, DenseNet, and InceptionV3 were utilized as pre-trained feature extractors, and the optimized selected features were classified by SVM, K-nearest neighbors (KNN), logistic regression (LR), and Naive bayes (NB). A Kaggle dataset of 10,000 colon histopathological images (5,000 benign and 5,000 malignant) was divided into training and independent test sets, and performance was further validated through 5-fold cross-validation. Results demonstrate that SqueezeNet + WOA + SVM hybrid achieved the highest accuracy (99.03%), precision (99.29%), and F1 score (99.03%), surpassing fine-tuned deeper models such as EfficientNet and Vision Transformer in this dataset. Statistical significance tests (p < 0.01) confirmed robustness.