Modellevaluation

Modellevaluation refers to the process of assessing how well a statistical or machine learning model makes predictions on data. The goal is to estimate performance on unseen data, compare different models, and inform decisions about deployment, monitoring, and improvement. Evaluation should align with the task objective, such as predicting a continuous quantity, a class label, or a probability.

Common steps include defining an evaluation objective, selecting appropriate metrics, and partitioning data into training, validation, and test sets. Cross‑validation, including k-fold or stratified variants, provides more reliable performance estimates than a single split. It is essential to avoid data leakage where information from the test set influences training, and to respect temporal ordering in time-series data.

Task-specific metrics commonly used in evaluation include: for classification, accuracy, precision, recall, F1, ROC‑AUC, and calibrated scores such as the Brier score; for regression, mean squared error, root mean squared error, mean absolute error, and R-squared; for ranking or probability estimation, area under the ROC curve and precision‑recall curves. Clustering and similarity tasks may use metrics such as silhouette score or adjusted Rand index. Calibration, interpretability, and robustness considerations are often evaluated alongside raw predictive performance.

Beyond metrics, evaluation addresses the bias–variance tradeoff, robustness to noise and data shifts, and reproducibility. After deployment, ongoing evaluation through monitoring and online experimentation (for example, A/B tests) helps detect drift and guide retraining to maintain performance over time.