Use of data mining techniques to determine presence of coronary artery disease and deriving a risk score by employing risk factors

This study focuses primarily on the problems of collaborative classification with missing data on Coronary Artery Disease (CAD) by applying machine learning algorithms and proposes a risk score prediction system consisting of a 4-classes dataset. Three imputation methods are applied: K-means, multilayer perceptron (MLP), and self-organizing maps (SOMs). The MLP imputation method is obviously the best method among those investigated with the metric values for sensitivity (0.90), and for specificity (0.18). Dataset imputed with MLP method is employed by transforming into a 4-classes structure. Using the feature selection and the sampling methods with the NN substantially improves the evaluation metrics. The results before the pre-process operations were detected as follows; 72.3% accuracy; after the operations, 84.1% accuracy were achieved with 0.84 sensitivity 0.94 specificity. This study also presents a hybrid classification procedure that uses Support Vector Machine (SVM) with LR on two distinctive datasets. The results show that the hybrid approach allows developing an efficient algorithm, which solves the problem with all imbalanced dataset training at one time.