| Resumen |
Machine learning systems allow solving a wide variety of problems present both in industry and in everyday life. The increasing available information has made possible the training of these models and reaches a precision that surpasses the performance of a human being when carrying out the same task. However, the large amount of information necessary to carry out the training task of these models is not always available or difficult to obtain. That is why several methods are used to increase that information from existing training data; these methods are called data augmentation methods. The use of data augmentation techniques allows increment artificially the numbers of samples of a small training dataset, improving the performance of a machine learning model in tasks for which the information available is limited. This paper presents a novel data augmentation technique based on explainable artificial intelligence, where new training samples are created from the explanations generated using the explainability method described in [1]. This method generates explanations based on the classification of useful regions found in an image. Using this method, we can improve the model performance and its accuracy on the test dataset. |
