Development of a novel hybrid approach based on convolutional neural networks-metaheuristic algorithms for image classification

Monkeypox, previously endemic to Africa and resembling smallpox, is a zoonotic disease that has recently garnered attention due to increased transmission among humans and outbreaks beyond Africa. While the disease typically self-limits, it can result in up to 10% mortality and severe complications. Therefore, it is important to develop effective strategies for detection, treatment, and prevention. Deep learning, particularly prevalent in medical applications requiring high accuracy and precision, has seen significant success in various complex problem-solving tasks. Deep learning-based approaches in image classification have particularly excelled, serving as decision support mechanisms in numerous healthcare challenges. In this study, a suitable dataset was utilized to classify images of individuals infected with Monkeypox and normal body images. The dataset comprised a total of 1139 images, including 587 Monkeypox and 552 normal images. Convolutional Neural Networks (CNNs) were employed for feature extraction, with pre-trained models such as AlexNet, DarkNet19, InceptionNet-v3, and ResNet18 preferred. Subsequently, the extracted features were classified using various classifiers, including Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Decision Tree. Following this, a novel hybrid approach integrating metaheuristic algorithms, specifically the Sea-horse Optimizer (SHO), was employed for feature selection. This process aimed to eliminate features that could adversely affect classification performance and select more effective ones. Experimental results demonstrate a significant enhancement in classification performance through the application of SHO for feature selection on features extracted from CNN models.