The last few decades have seen an increasing trend in the incidence of melanoma – a malignant skin tumor. Given the results of the projections being made for the upcoming one-year period, it seems that this trend will continue. The most important factor on which the successful treatment of this disease depends is its early and timely detection. In order to increase the success of correct detection and prevent further development of the disease, machine learning models whose intention is to automate the detection of malignant tumors and their classification based on characteristics such as shape, color and size are being developed. Convolutional neural networks are being applied in applications that involve image processing, such as the classification process. The main goal of this paper is to improve the early detection of malignant irregularities found on the skin using automatic, computer-assisted diagnostics, specifically, by applying the functionalities of neural networks in the problems of classifying images containing malignant changes in order to increase treatment success. Furthermore, the existing approaches to the classification of skin irregularities are presented and the performance of the implemented systems is analyzed.