Support vector machines and genetic algorithms for detecting unstable angina

In this communication we present a combination of two state-of-the-art. machine learning methods for predicting mortality in patients with unstable angina (UA). Support Vector Machines (SVM) are used as non-linear discrimination tools. However, before building the models, selection of the best subset of variables is carried out with Genetic Algorithms (GA). The best subset of descriptors selected - by the GA was constituted by five variables from the originally 75 collected. The data was split into 4 training set (483 patients; 22 cases with UA) - and a validation set (243 patients; 12 of cases with UA). The criterion used to select the best model was based on the sensitivity (SE); specificity (SP) and negative predictive values (NPV) in the validation data set. The final SVM model (RBF kernel) yielded good results (SE = 66.67%, SP = 79.77% in the validation set). The recognition rate was 79.12% and a high rate of NPV (97.87%) was obtained. Methods proposed have proven to be well-suited for this problem, simplifying the solution and providing excellent discrimination scores.