threcv: Optimize threshold on posterior probability
In jinyung/otolith: Identification of fish otolith using digital images

Description Usage Arguments Details Value References See Also

To estimate the effect of setting a threshold on posterior probability to reject "unsure" predictions.

1	threcv(X, Y, saveplot = FALSE, plotsize = 1000, run = 30, k = 5, level)

`X`	matrix of PC scores from `rGPA` object (the `score` value)
`Y`	a factor giving the grouping, e.g. the `sp` value from `routine1` object or obtain from `getclass`
`saveplot`	logical. The plot will be saved if `TRUE`, and no plot is displayed in the window. Note that outliers of boxplot are not plotted.
`plotsize`	numeric. Plot size for the plot saved into file (used only when `saveplot=TRUE`), unit in pixel.
`run`	argument to be passed to `mrkfcv`
`k`	argument to be passed to `mrkfcv`
`level`	numeric vector giving the threshold range to be tested. see details.

only lda is supported now. lda decide the class (e.g. species) based on the calculated posterior probability. New sample is assigned to the class that has the higher posterior probability. We can reject the classification result if the posterior probability is low (the "unsure" classification). In this way we can increase the overall accuracy (correct predictions out of all the reported results), but the tradeoff is the decrease in number of reported classification result. The function test the threshold value of 0.50 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95 0.99 if level is not provided. Since thresholding is just an option, the choice of threshold value is really arbitrary and depends on one's objectives.

a matrix and a plot are returned, giving the overall accuracy and the total reported prediction percentage over a range of threshold values.

Beleites, C., & Salzer, R. (2008). Assessing and improving the stability of chemometric models in small sample size situations. Analytical and Bioanalytical Chemistry, 390(5), 1261-1271.

Similar: pccv, harcv

Which this function wraps: mrkfcv

jinyung/otolith documentation built on May 19, 2019, 10:36 a.m.