F1 Score

The F1 Score combines precision and recallPrecision and RecallPrecision and Recall are complementary classification metrics that measure prediction exactness and completeness resp... into a single metric using the harmonic mean: F1 = 2 * (precision * recall) / (precision + recall). Unlike the arithmetic mean, the harmonic mean penalizes extreme imbalances, so a high F1 score requires both precision and recall to be reasonably high. An F1 of 1.0 indicates perfect precision and recall, while an F1 near 0 indicates poor performance on at least one metric. The F1 score is particularly useful when there is no predetermined preference between precision and recall and when class distributions are imbalanced. F1 Score in Geospatial Model EvaluationThe F1 score is widely used to evaluate geospatial classification and detection models. Per-class F1 scores reveal which land cover types or object categories are well-classified and which need improvement. Macro F1 averages per-class F1 scores equally, giving equal weight to rare and common classes, which is important in geospatial applications where minority classes like wetlands or specific infrastructure types are often the most important to detect. Micro F1 computes F1 from aggregated counts, effectively weighting by class frequency. For object detectionObject DetectionObject Detection is a computer vision technique that identifies and localizes specific objects within images or video..., F1 at specific IoU thresholds combines detection quality with localization accuracy. When to Use F1 and AlternativesThe F1 score is most valuable when both false positives and false negatives carry similar costs, and when class imbalance makes accuracy misleading. In geospatial applications where one error type is more costly, the F-beta score generalizes F1 by introducing a parameter beta that controls the precision-recall tradeoff. F2 emphasizes recall (useful for damage detection), while F0.5 emphasizes precision (useful for high-confidence mapping). Reporting F1 alongside individual precision and recall provides the most complete picture of classification performance.