Detecting skin cancer was earlier a critical task for successful treatment and recovery. Unfortunately, traditional methods of diagnosis rely heavily on the expertise of dermatologists, whose availability can be limited. This paper proposed a novel approach using weight classification to balance the various skin cancer groupings and adjust for data anomalies. On the Skin cancer MNIST dataset to identify seven different types of skin lesions as skin cancer. A Convolutional neural network was designed to predict the type of cancerous and proposed model is further trained using a range of hyperparameter combinations to improve its accuracy. In summary, this research presents an innovative approach to skin cancer detection using CNN. After performing all the necessary preprocessing and training steps the proposed framework offers a promising solution to early skin cancer detection with an accuracy of 91.75% results.