Easy Predictive Analytics provides a Model Level as a reference for how good a prediction model created from prediction model creation (training) data is.

The model level is expressed as 1 to 5 stars, and the higher the number of stars, the better the prediction model.

The model level is calculated based on the evaluation metrics that are presented together on the model summary screen. The accuracy evaluation from which the calculation is made depends on the problem as follows:

• For binary classification：AUC
• For multiclass classification: percentage of correct answers (Accuracy)
• numerical prediction: Coefficient of Determination
• For time series predictions: Median Absolute Error Ratio

In general, the higher the model level, the more accurate your prediction model will be. However, each use case must determine whether the model is actually a useful model.