Thresholding Classifiers to Maximize F1 Score
This addresses the challenge of optimizing F1 scores for imbalanced data in classification tasks, providing theoretical insights with practical applications in domains like medical document labeling, though it is incremental as it builds on existing thresholding methods.
The paper tackles the problem of maximizing F1 scores in binary and multilabel classification by deriving the optimal decision threshold for classifiers with real-valued outputs, showing that for well-calibrated probabilities, the optimal threshold is half the optimal F1 score, and applies this to predict 26,853 labels for Medline documents.
This paper provides new insight into maximizing F1 scores in the context of binary classification and also in the context of multilabel classification. The harmonic mean of precision and recall, F1 score is widely used to measure the success of a binary classifier when one class is rare. Micro average, macro average, and per instance average F1 scores are used in multilabel classification. For any classifier that produces a real-valued output, we derive the relationship between the best achievable F1 score and the decision-making threshold that achieves this optimum. As a special case, if the classifier outputs are well-calibrated conditional probabilities, then the optimal threshold is half the optimal F1 score. As another special case, if the classifier is completely uninformative, then the optimal behavior is to classify all examples as positive. Since the actual prevalence of positive examples typically is low, this behavior can be considered undesirable. As a case study, we discuss the results, which can be surprising, of applying this procedure when predicting 26,853 labels for Medline documents.