Model performance assessment
Once a prediction model is developed then it needs to be validated to
see or quantify how good the predictions from the models are, often
referred to as model performance. There are different methods and
metrics to assess the performance of a prediction model. These methods
and metrics depend on the type of modeling technique used in model
developing which again largely depends on the outcome of interest. We
will restrict our discussion of model performance assessment for binary
or survival outcomes, common in health research. For binary and survival
outcomes, the most commonly used measures include the Brier score to
indicate overall model performance, the concordance statistic (also
known as the C-statistic) for discriminative ability, and
goodness-of-fit statistics for calibration.