Negative Log Likelihood (NLL) is a statistical measure used to evaluate the performance of a probabilistic model. It quantifies how well a given model predicts a set of beobachtete Daten points. In essence, NLL assesses the model’s ability to assign high probabilities to the actual outcomes in a dataset.
Mathematisch ist die Log-Wahrscheinlichkeit der Logarithmus der Likelihood-Funktion, which describes the probability of the observed data given a set of parameters. The likelihood function measures how likely the observed data is under different parameter values. By taking the logarithm, we transform products into sums, making calculations more manageable and numerically stable. The NLL is defined as the negative of this log likelihood. This transformation is useful because Optimierungsalgorithmen often seek to minimize a Verlustfunktion; thus, minimizing NLL corresponds to maximizing the likelihood of the observed data under the model.
NLL is commonly used in various fields, including machine learning, statistics, and künstliche Intelligenz, particularly in tasks involving classification and regression. For example, in binary classification problems, NLL can be employed to evaluate how well a logistic regression model predicts the probability of class membership. Lower values of NLL indicate better model performance, as they signify that the model is assigning higher probabilities to the correct outcomes.
However, while NLL is a powerful tool, it is essential to be cautious of overfitting, where a model performs well on training data but poorly on unseen data. Techniques such as regularization kann zusammen mit NLL eingesetzt werden, um dieses Risiko zu verringern.