Overfitting (EN)
ConceptExcessive model adaptation to training data with poor generalization
Overfitting in the context of machine learning
Overfitting occurs when a machine learning model is too strongly adapted to the training data and cannot generalize well to new, unseen data. The model not only learns the underlying patterns but also random noise or outlier data in the training data. This leads to high accuracy on the training data but poor performance on test or real-world data. Regularization techniques and cross-validation help prevent overfitting.
Visualization of the overfitting problem
flowchart TD A[Datenpunkte] --> B[Modelltraining] B --> C[Überangepasstes Modell] B --> D[Gut generalisierbares Modell] C --> E[Hohere Genauigkeit auf Trainingsdaten] C --> F[Niedrigere Genauigkeit auf Testdaten] D --> G[Mäßige Genauigkeit auf Trainingsdaten] D --> H[Gute Genauigkeit auf Testdaten] style E fill:#f9f,stroke:#333,stroke-width:2px style F fill:#f99,stroke:#333,stroke-width:2px style G fill:#9f9,stroke:#333,stroke-width:2px style H fill:#9f9,stroke:#333,stroke-width:2px
Im Kontext
- Wird typischerweise zusammen mit Regularisierungstechniken wie L1/L2-Regularisierung, Dropout und Early Stopping verwendet
- Verwandt zu: Underfitting, Bias-Variance-Tradeoff, Cross-Validation, Generalisierung
- Beispiel-Einsatz: Ein n