Calibrating kmeans

This function allows to calibrate the kmeans algorithm. It is possible to obtain the best k value and the best embedded algorithm in kmeans.

calkmeans(
  img,
  k = NULL,
  iter.max = 10,
  algo = c("Hartigan-Wong", "Lloyd", "Forgy", "MacQueen"),
  iter = 30,
  verbose = FALSE,
  ...
)

Arguments

img

RasterStack or RasterBrick.

k

This argument is NULL when the objective is to obtain the best k value. If the objective is to select the best algorithm embedded in kmeans, please specify a k value.

iter.max

The maximum number of iterations allowed. See kmeans.

algo

It can be "Hartigan-Wong", "Lloyd", "Forgy" or "MacQueen". See kmeans.

iter

Iterations number to obtain the best k value. iter must be greater than the number of classes expected to be obtained in the classification. Default is 30.

verbose

This parameter is Logical. It Prints progress messages during execution.

...

Options to be passed to the function. See 'Details'.

Details

If the idea is to find the optimal value of k (clusters or classes), k = NULL as an argument of the function must be put, because the function find k for which the intra-class inertia is stabilized. If the k value is known and the idea is to find the best algorithm embedded in kmeans (that maximizes inter-class distances), k = n, which n is a specific class number, must be put.

If warnings such as "Quick-TRANSfer stage steps exceeded maximum" or "did not converge in 10 iterations" are obtained, it will be necessary to increase the iterations in 20 or 30 (i.e., inter.max = 20 or iter.max = 30). This issue is usually obtained with "Hartigan-Wong". See details of kmeans.

References

Tarazona, Y., Maria, Miyasiro-Lopez. (2020). Monitoring tropical forest degradation using remote sensing. Challenges and opportunities in the Madre de Dios region, Peru. Remote Sensing Applications: Society and Environment, 19, 100337.

Gareth James, Daniela Witten, Trevor Hastie, Robert Tibshirani. (2013). An introduction to statistical learning : with applications in R. New York: Springer.

Examples

library(ForesToolboxRS)

# Load the dataset
data(img_l8)

# Select the best k value
best_k <- calkmeans(
  img = img_l8[[1:2]], k = NULL, iter.max = 10,
  algo = c("Hartigan-Wong", "Lloyd", "Forgy", "MacQueen"), iter = 30
)
# Jambu Elbow
plot(best_k)


# Select the best embedded algorithm in kmeans
best_algo <- calkmeans(
  img = img_l8[[1:2]], k = 4, iter.max = 10,
  algo = c("Hartigan-Wong", "Lloyd", "Forgy", "MacQueen"), iter = 30
)

# Choose the algorithm with the highest value
best_algo