Logförlusten

Logförlusten is a Swedish term that translates to "log loss" in English. It is a performance metric used in machine learning, particularly for classification tasks. Log loss quantifies the uncertainty of a prediction by penalizing incorrect predictions. The lower the log loss, the better the model's predictions are. It measures the accuracy of a classifier where the prediction is a probability value between 0 and 1.

The formula for log loss involves the true label (0 or 1) and the predicted probability for that label. If the true label is 1 and the model predicts a probability close to 1, the log loss will be low. Conversely, if the model predicts a probability close to 0 for a true label of 1, the log loss will be very high, indicating a poor prediction. The same logic applies if the true label is 0. Log loss is a differentiable loss function, making it suitable for gradient-based optimization algorithms used in training machine learning models.

In practice, log loss is often used to evaluate the performance of models like logistic regression, support vector machines with probability estimates, and neural networks trained for classification. It provides a more nuanced evaluation than simple accuracy, especially when dealing with imbalanced datasets or when the confidence of the predictions is important. A model with a high log loss suggests that its probability estimates are poorly calibrated.