Comparing classification methods for predicting survival probabilities in the elderly

We examine how different classification methods and different factors affect the prediction of the survival probabilities in the elderly. We group health related attributes into different categories and generate different frailty indexes from them. We explore how these different indexes, weighted and un-weighted affect the classification. All experiments have been cross-validated using 10-fold evaluation. The two performance metrics used are classification accuracy and area under the ROC curve. Our approach performs better than the frailty index model, but not as good as the ANN method when AUC is used.