In recent years, Malignant Melanoma Cancer has caused an increased exponential in human diseases, for this reason, it is essential to detect it from its early stages. Deep Learning is one of the most applied technologies for the analysis of images oriented to medicine, facilitating the diagnosis of diseases in patients, allowing them to make accurate decisions about their health. In this paper, we propose a convolutional neural network architecture derived from the evaluation of different convolutional neural networks that meet the objective of obtaining more pressure on the information of the acquired image. The model for the problem is based on a binary distribution, 1 in case of malignant and 0 for benign, so that melanoma can be detected early and is very useful, for this we used 2 different datasets with a total of 2650 images for training the architecture. Finally, a comparison of the results obtained in other research has been made, where the metrics of our project are considerably improved by having 3 layers. This new architecture is a proposed solution for the optimization of training and validation of images.