Skin Lesion Classification using Deep Feature Fusion and Selection Using XGBoost Classifier

Skin cancer is a potentially fatal condition that needs to be detected as soon as possible in order to be treated effectively. Deep convolutional neural networks (DCNNs) have shown promising results in the prediction of skin cancer in recent years. This study presents a novel approach for skin cancer identification using deep feature fusion and selection based on the significance score obtained with the XGBoost classifier. The proposed method combines features from the state-of-the-art pre-trained DCNNs, such as EfficientNetB3, ResNet50, VGG16, ConvNeXtTiny, and DenseNet121, to extract high-level features from dermoscopic images. These features capture the intricate patterns and textures associated with malignant and benign skin cancers. Based on the relevance score that the XGBoost classifier awarded to each feature, the K-Best (K=1000) features were chosen. Using the XGBoost classifier, the suggested technique has successfully classified dermoscopic pictures of benign and malignant melanoma, yielding an area under the curve value of 0.95. In comparison to stand-alone DCNN-based techniques, the experimental findings show that the suggested feature fusion and selection strategy has obtained greater accuracy. Additionally, an analysis has been done on how the performance of the suggested classifier is affected by the quantity of characteristics that are chosen.