PMC:4289553 / 20273-21086 JSONTXT

{"target":"http://pubannotation.org/docs/sourcedb/PMC/sourceid/4289553","sourcedb":"PMC","sourceid":"4289553","source_url":"https://www.ncbi.nlm.nih.gov/pmc/4289553","text":"While RF and LR models consistently performed well, CART consistently performed poorly. The poor performance of CART modelling may be explained by the fact that continuous variables need to be categorized, with optimal cut-offs determined from all possible cut-off points, and that possibly unnecessary higher-order interactions are assumed between all predictor variables. RF modelling is an obvious improvement over CART modelling [24]. It is hence remarkable that CART is still advocated as the preferred modelling technique for prediction in some disease areas, such as trauma [25]. A researcher must always carefully consider which modelling technique is appropriate in a specific situation. Using, for instance, a random forest technique just because the number of subjects is over 10,000 is too simplistic.","tracks":[]}