classAccuracy: classAccuracy

In AndreasChristianHill/classoptimr: Optimal classification schemes for prediction models with continuous response variable

Description

Calculates common classification accuracies measures known from remote sensing applications based on a Confusion Matrix that compares Reference classes to their respective class predictions. The computed accuracy measures are:

• Overall Classification Accuracy (OAA)

• Producer's Accuracy (PA)

• User's Accuracy (UA)

• Cohen's Kappa Coefficient (k)

• Quantity Disagreement

• Allocation Disagreement

Usage

classAccuracy(..., conf.level = 0.95)

## S3 method for class 'hsmclass'
classAccuracy(object, conf.level = 0.95, ...)

## Default S3 method:
classAccuracy(refdata, predictions, equal.int = NA,
  def.int = NA, conf.level = 0.95, ...)

Arguments

...	further arguments passed to or used by methods.
conf.level	the confidence level used to compute the confidence intervals of the overall accuracy.
object	object of class 'hsmclass' created by function HSMclass.
refdata	vector containing the values of the continuous response variable used in the prediction model.
predictions	vector containing the predictions for the response values of the prediction model. Note:(refdata and predictions have to correspond to each other).
equal.int	an equidistant class interval to be evaluated as classifciation scheme. Defaults to NA. Can only be set if def.int is NA.
def.int	a vector defining an arbitrary set of class breaks to be evaluated as classifciation scheme. Defaults to NA. Can only be set if def.int is NA.

Value

classAccuracy returns an object of class "classaccur"

An object of class "classaccur" returns a list of the following components:

rsquared_val

the coefficient of determination of the prediction model, calulated based on refdata and predictions

.

predictions	vector containing the predictions given to classAccuracy
classwidth	vector containing the class width
def.classbreaks	vector containing arbitrary class break values
equal.classbreaks	vector containing equidistant class break values
overall.accuracy	the overall accuracy of the classification scheme
conf.oaa	vector containing [1] the lower confidence limit of the OAA, [2] the upper confidence limit of the OAA, and [3] the confidence level
prodaccuracy	a matrix containing the producer's accuracy for each class
usersaccuracy	a matrix containing the user's accuracy for each class
no.ref.classes	a vector containing the number of reference data for each class
cohenskappa	Cohen's Kappa Coefficient
map.accuracy	the map accuracy
Quantity.Disagreement	the quantity disagreement
Allocation.Disagreement	the allocation disagreement

References

Congalton, R.G.; Green, K. Assessing the Accuracy of Remotely Sensed Data: Principles and Practices; Lewis Publications: Boca Raton, FL, USA, 1999; p. 137.

Richards, J.A. Remote Sensing Digital Image Analysis: An Introduction, 5th ed.; Springer: Berlin, Germany, 2013.

Hill, A., Breschan, J., & Mandallaz, D. (2014). Accuracy assessment of timber volume maps using forest inventory data and LiDAR canopy height models. Forests, 5(9), 2253-2275.

Examples

#----
# 1.) Example: classification accuracy for equidistant class width of 100:
#----

acc.equal<- classAccuracy(refdata.gr, predictions.gr, equal.int = 100)
summary(acc.equal)


#----
# 2.) Example: classification accuracy for arbitrary class breaks:
#----

acc.def<- classAccuracy(refdata.gr, predictions.gr,
                       def.int = c(0, 150, 200, 430, 610, 880))
summary(acc.def)


#----
# 3.) Example: classification accuracy for optimal class breaks:
#----

# run HSMclass:
## Not run: 
hsm<- HSMclass(refdata.gr, predictions.gr, nclasses = 6,
              iterations = 1000, coolfactor=0.99, InitTemp = 80,
              weight.norefs = 2, weight.classwidth = 2)

# calculate accuracy:
acc.opti<- classAccuracy(hsm)
summary(acc.opti)

## End(Not run)

AndreasChristianHill/classoptimr documentation built on May 29, 2019, 12:23 p.m.