Learning from machine learning mistakes - KDnuggets

When we analyze machine learning model performance, we often focus on a single quality metric. With regression problems, this can be MAE, MAPE, RMSE, or whatever fits the problem domain best. Optimizing for a single metric absolutely makes sense during training experiments. This way, we can compare different model runs and can choose the best one. But when it comes to solving a real business problem and putting the model into production, we might need to know a bit more.