optim.thresh {SDMTools} | R Documentation |
optim.thresh
estimates optimal threshold values
given eight methods.
Note: this method will
exclude any missing data.
optim.thresh(obs, pred, threshold = 101)
obs |
a vector of observed values which must be 0 for absences and 1 for occurrences |
pred |
a vector of the same length as |
threshold |
a single integer value representing the number of equal interval threshold values between 0 & 1 |
Returns a list of the optimal thresholds for the different
methods. If the list item is a single value, that is the
optimal threshold but if two values are reported for the
method, this represents the range in thresholds that are
equal for that threshold selection method.
The
returned list includes the single or range in thresholds
selected using the following methods:
min.occurence.prediction |
is the minimum prediction for the occurrence (presence) records |
mean.occurence.prediction |
is the mean prediction for the occurrence (presence) records |
'10.percent.omission' |
is the threshold value or range in values that excludes approx. 10 percent of the occurrence records |
'sensitivity=specificity' |
is the threshold value or range in values where sensitivity is equal to sensitivity |
'max.sensitivity+specificity' |
is the threshold value or range in values that maximizes sensitivity plus specificity |
maxKappa |
is the threshold value or range in values with the maximum Kappa statistic |
max.prop.correct |
is the threshold value or range in values with the maximum proportion of presence and absence records correctly identified |
min.ROC.plot.distance |
is the threshold value or range in values where the ROC curve is closest to point (0,1) (or perfect fit) |
Jeremy VanDerWal jjvanderwal@gmail.com
accuracy
, auc
,
Kappa
, omission
,
sensitivity
, specificity
,
prop.correct
, confusion.matrix
#create some data obs = c(sample(c(0,1),20,replace=TRUE),NA); obs = obs[order(obs)] pred = runif(length(obs),0,1); pred = pred[order(pred)] #calculate the optimal thresholds optim.thresh(obs,pred)