Metric Calculations
Accuracy
\({tp + tn} \over {p + n}\)Accuracy measures how many correct instances were predicted divided by total number of instances.
Balanced Accuracy
\({1 \over 2}{({tp + tn}) \over ({p + n})}\)Calculates average of proportion of correctly predicted classes.
Geometric Mean
\(({tp \over p} * {tn \over n})^{1 \over 2}\)Product of true positive over positive instances and true negatives over negative instance inversed.
Precision
\({tp} \over {p}^¬\)Indicates the proportion of the data points that are relevant.
Recall
\({tp} \over {p}\)Indicates the proportion of the data points that are relevant.
F1-Score
\({2*{pre * rec} \over {pre + rec}}\)Weighted average of Precision and Recall.
Gilbert's Skill Score
\({tp - r} \over {tp + fp + fn - r}\)Doolittle's Skill Score
\(({tp * tn - fp * fn})^2 \over {(tp + fp) * (tp + fn) * p * n}\)True Skill Statistic
\({tp \over p} - {fp \over n}\)Represents the matches and also mismatches between observatiosn and predictions.
Heidke Skill Score
\({2*((tp * tn) - (fn * fp))} \over {p * (fn + tn) + n * (tp + fp)}\)A measure of skill in forecasts. Finds the correct number of forecasts.
Youden Index
\({tp * tn - fn * fp} \over {(tp + fn) * (fp + tn)}\)Captures performance of a dichotomous diagnostic test.