Introduction: The identification and monitoring of benign moles and skin cancers leads to a challenging task because of the usual standard significant skin patches. Actually, the skin lesions vary very little in their look and only limited amount of information is available. There are seven fundamental types of skin cancer like Basal Cell Carcinoma (BCC), Melanoma and Squamous Cell Carcinoma (SCC) whereas Melanoma is the highly risky which has low survival rate. Objective: This work classifies skin lesions with the help of Convolution Neural Network and the images are trained end-to-end. A dataset comprised of 10000 clinical images were trained using Convolution Neural Network (CNN). Materials and Methods: The skin cancer identification process is generally separated into two basic components, image pre-processing which includes classification of images and removing the duplicate images and sharpening, which resizes the skin image. This work discusses a methodology to segment the high-level skin lesion and identification of malignancy more accurately with the help of deep learning: 1) Construction of a neural network, which detects the edge of a huge lesion accurately; 2) Designing model that can run on mobile phones. The model designed a transfer learning which is based deep on neural network and the fine turning that supports to attain high prediction accuracy. Results: The dataset comprises of a total of 10,000 images stored in two folders. The information about the data is stored in a data frame. Total 10000 dermoscopic images contains 374 melanoma images, 254 seborrheic keratosis images and 1372 nevus images. Using transfer learning validation loss, Top-2 accuracy and Top-3 accuracy have been calculated. The result has been compared with the different models. Conclusions: The proposed system can categorize healthy skin lesions, eczema, acne, malignant and benign skin lesions. The proposed work investigates the attributes acquired by the deep convolutional neural network. The attributes are extracted and the datasets were divided into seven different categories. Based on that categories the data was trained and validated. Based on the calculation the validation loss, top-2 accuracy, top-3 accuracy was calculated.