Overfitting / Underfitting
Refers to a ML model not being able to generalize to new data.
Measures
- Dataset Splitting
- K-Fold Cross Validation
- Nur Essentielle Features verwenden
- Anpassung Evaluation / Loss Funktion
Regularization
- L1
- L2
- Early Stopping
- Dropout: At every training step, ignore a number (%) of neurons.