misclassifikation

Misclassification is the incorrect assignment of an observation to a category or class. In statistics, data science, and related fields, it denotes errors in labeling or prediction.

In supervised learning, a misclassification occurs when the model's predicted label does not match the true label. Evaluation commonly uses a confusion matrix with true positives, false positives, true negatives, and false negatives. The misclassification rate equals (FP + FN) / N, while accuracy equals (TP + TN) / N. These measures summarize the frequency and distribution of errors.

Causes include noisy or mislabeled training data, ambiguous examples, inadequate feature representation, limited model capacity, class imbalance, and data drift. Misclassification can be random (noise) or systematic (bias). Differential misclassification refers to when the probability of mislabeling differs by true class, potentially biasing analyses, while non-differential misclassification tends to affect all classes more uniformly.

Implications include biased performance estimates, degraded decision quality, and fairness concerns. Metrics such as precision, recall, F1 score, and the area under the ROC curve complement accuracy by revealing where misclassifications occur and how severe they are for different classes or thresholds.

Mitigation strategies center on data quality and model design. They include careful labeling and adjudication, using multiple annotators, data cleaning, addressing class imbalance, selecting appropriate models, calibrating probabilistic outputs, threshold tuning, and applying active learning to prioritize uncertain samples for labeling. In epidemiology and related fields, misclassification bias is a recognized concern that can distort effect estimates if disease or exposure status is misclassified.