DEVELOPMENT OF AN INTELLIGENT DEVICE FOR SKIN DISEASE MONITORING

Abstract

This article aimed to develop an intelligent algorithm based on convolutional neural networks for the automated detection of melanoma, the most lethal type of skin cancer. A public dermatoscopic image dataset from Kaggle was used. The model was trained using data augmentation techniques and efficient preprocessing to improve generalization and reduce overfitting. The methodology involved the use of Python language supported by the Keras platform, and the experiment was conducted in the Google Colab environment using GPU. The dataset was split into 80% for training and 20% for testing. The results showed a promising performance, achieving an accuracy of 89.09%, sensitivity of 85.7%, precision of 91.3%, and an F1-score of 88.4%. It is concluded that the developed model can be an effective auxiliary tool for the early diagnosis of melanoma, especially in contexts with limited access to specialists, contributing to more accessible and accurate medical care