G-mean method

Calculate the optimal probability classification threshold using the G-Mean method.

Usage

calc_gmean(estimates, truth)

Arguments

estimates

A vector of classification probabilities. Values should represent the probability of 1 in the truth argument.

truth

An integer vector of 0 and 1 representing the true classifications.

Value

A numeric scalar representing the optimal probability threshold.

Details

The G-mean method (Kubat & Matwin, 1997) is defined as the square root of the product of sensitivity and specificity at a given threshold. The optimal threshold is the threshold with the greatest g-mean.

The optimality criterion is then defined as:

$$\text{max}\sqrt{(sensitivity * specificity)}$$

References

Kubat, M. & Matwin, S. (1997, July 8-12). Addressing the curse of imbalanced training sets: One-sided selection [Paper presentation]. International Conference on Machine Learning, Nashville, TN.

See also

Other optimal threshold methods: calc_cz(), calc_topleft(), calc_youden()

Examples

calc_gmean(estimates = dcm_probs$att1$estimate,
            truth = dcm_probs$att1$truth)
#> [1] 0.3170266

calc_gmean(estimates = dcm_probs$att2$estimate,
            truth = dcm_probs$att2$truth)
#> [1] 0.4363361

calc_gmean(estimates = dcm_probs$att3$estimate,
            truth = dcm_probs$att3$truth)
#> [1] 0.3722605