When creating a predictive model, data miners need to “tune” it to our client's needs. We need to strike the right balance for them between missed opportunities and false alarms. For some clients, the model's cut-off point between 'promising' and 'unpromising' must be set lower to avoid missed opportunities. For others it needs to be set higher to avoid false alarms. Since no predictive model can be perfect, we need to tweak it to make the "right kind of mistakes." Read the full article.