| True condition | Model performance statistics | ||
---|---|---|---|---|
Positive (P) = Dead | Negative (N) = Live | |||
Predicted condition | Positive (P) = Dead | True Positive (TP) Dead trees that were predicted to be dead | False Positive (FP) Live trees that were predicted to be dead | Positive predictive value (PPV) \( \boldsymbol{PPV}=\frac{\boldsymbol{TP}}{\boldsymbol{TP}+\boldsymbol{FP}} \) Dead trees that were predicted to be dead divided by predicted dead trees Example use: Prescribed fire planning where there is a need to kill small shade-tolerant trees to reduce future fire risk. |
Negative (N) = Live | False Negative (FN) Dead trees that were predicted to be live | True Negative (TN) Live trees that were predicted to be live | Negative predictive value (NPV) \( \boldsymbol{NPV}=\frac{\boldsymbol{TN}}{\boldsymbol{TN}+\boldsymbol{FN}} \) Live trees that were predicted to be live divided by predicted live trees Example use: Prescribed fire planning where there is a need to avoid killing large or old trees. | |
Model performance statistics | Sensitivity (Sens) \( \boldsymbol{Sens}=\frac{\boldsymbol{TP}}{\boldsymbol{TP}+\boldsymbol{FN}} \) Dead trees that were predicted to be dead divided by total dead trees Example use: Post-fire salvage in campground, where there is a need to remove any trees that may die and become a hazard. | Specificity (Spec) \( \boldsymbol{Spec}=\frac{\boldsymbol{TN}}{\boldsymbol{TN}+\boldsymbol{FP}} \) Live trees that were predicted to be live divided by total live trees Example use: Post-fire salvage where there is a need to avoid harvesting large trees that may survive (e.g., potential seed trees or large wildlife trees.) | Accuracy (ACC) \( \boldsymbol{ACC}=\frac{\boldsymbol{TP}+\boldsymbol{TN}}{\boldsymbol{TP}+\boldsymbol{TN}+\boldsymbol{FP}+\boldsymbol{TN}} \) Correctly classified live and dead trees divided by total trees Example use: Need to optimize multiple objectives. |