<p>Identifying skin cancer manually from dermatoscopic images poses challenges due to the resemblance in appearance across various skin lesion classes. Applying machine learning techniques can provide faster and more accurate predictions that will assist physicians in making easy diagnoses. The class imbalance problem in the dataset is overcome with SMOTE-TOMEK and target-specific augmentation methods. Augmentation is done by rotating, flipping, shifting, zooming, and varying the brightness of the images. The ultimate effort is given to design an appropriate classifier that receives the raw images as input and accurately predicts. A homogeneous ensemble learning-based model with three support vector machines (EnsembleSVM) is designed to detect skin cancer. The ensemble model consists of two stages. In the first stage, two support vector machine (SVM) models are connected in parallel. One of the two base models is trained with the balanced data generated using SMOTE-TOMEK, whereas the second SVM model is trained with the augmented data. The final classification is done by another SVM model stacked to the ensemble of two SVMs. The proposed algorithm is verified using the HAM10000 dataset. The proposed method analyses the effect of the class imbalance problem and provides a solution. Appropriate choice of classifying model better, i.e., combining the models in an ensemble yielded a training accuracy of 99.9% and a test accuracy of 98.2%. The effectiveness of the suggested technique in identifying cancer from skin lesions is demonstrated.</p>