Skin Lesion Classifier: Skin cancer is recognized as the most common kind of cancer in the world. It could be deadly if not identified at the primary stage, which makes early detection very crucial. In this project, we present a novel approach for classifying seven types of skin lesions using an ensemble learning-based model with weighted averaging. The ensemble is constructed using five deep neural network models, namely ResNeXt, SeResNeXt, ResNet, Xception, and DenseNet. Our models are trained and evaluated on a dataset of 18,730 dermoscopy images from the HAM10000 and ISIC 2019 datasets, incorporating class balancing, noise removal, and data augmentation techniques. Through grid search optimization, we determined the best combination of base models, achieving impressive macro-average recall scores of 88%, 89%, 91%, 88%, and 84% for ResNeXt, SeResNeXt, DenseNet, Xception, and ResNet respectively. By employing a simple average ensemble, we further improved the performance to 93%, and the weighted average ensemble yielded an impressive 94% recall score. These findings highlight the balanced contribution of each base model in our final ensemble, as revealed by the grid search analysis. 

Paper Link 1: https://www.sciencedirect.com/science/article/pii/S2352914821001465

Paper Link 2: https://ieeexplore.ieee.org/abstract/document/9393155