Modelos de aprendizaje automático para el apoyo en la clasificación de tipos de cáncer a partir de datos estructurados y no estructurados de expedientes clínicos

Abstract

The existence of large volumes of data generated by the health area presents an important opportunity for analysis. This can obtain information to support physicians in the decisionmaking process for the diagnosis or treatment of diseases, such as cancer. The present work shows a methodology for the classification of patients with liver, lung and breast cancer, through machine learning models, to obtain the model that performs best in the classification. The methodology considers three classification models: Support Vector Machines (SVM), Multi-Layer Perceptron (MLP) and AdaBoost using both structured and unstructured information from the patient's clinical records. Results show that the best classification model is MLP using only unstructured data, obtaining 89% of precision, showing the usefulness of this type of data in the classification of cancer patients.