Prediction in the diagnosis of breast cancer tumors using classification methods

Abstract

The present work consists of the application of data sciences with the objective of predicting whether a breast cancer tumor is benign or not, for this the following classification method are used: neural networks, random forests and vector support machine. A dataset from the University of Wisconsin Hospital related to breast cancer was used. Confusion matrices are used to know the measurements of the forecast models and the ROC (Receiver Operating Characteristics) curve to determine the discriminant capacity of these, based on the value of AUC (Area Under the Curve). The proposed models reach accuracy values ​​that indicate that very accurate predictions can be made with them, although it is important to highlight that the vector support machine model is the most convenient to use since its level of accuracy in the forecast exceeds 99 %. The use of these techniques is recommended in hospitals and laboratories where the detection of this disease is carried out, since it can be a support tool in the diagnosis of breast cancer.