2

I have been reading this forum but cannot find anything specific enough to address my problem.

I have classified disease in the below image (red spots), and verified disease by GPS (Red Circles).

I believe I have these three types of variables: • True Positive (TP): Correctly classified as the class of interest (GPS Spatially matches areas of Classified Disease) • False Positive (FP): Incorrectly classified as the class of interest (No GPS Data for Classified Disease) • False Negative (FN): Incorrectly classified as not the class of interest (GPS Data but no red spot/classification).

Based on this, what is a simple way I can derive single classification accuracy for my entire image- considering both False Positives and Negatives- and are there weights I can apply to improve my accuracy? enter image description here

amoeba
  • 93,463
  • 28
  • 275
  • 317
Alexander
  • 75
  • 6
  • 1
    You seem to be discarding a large amount of information just to be able to do classification, as opposed to using probabilistic thinking which allows for gray zones and much more. – Frank Harrell Jan 04 '17 at 14:16
  • Hi Frank! It may seem that way- but I am showing just a small portion of an image here- and all I want to know is how accurately my classification technique can detect known areas of disease verified by GPS. – Alexander Jan 04 '17 at 14:26
  • 1
    The fact that you are using a classification technique is part of the problem unless the problem is deterministic, e.g., no unit-to-unit variation, no measurement error. – Frank Harrell Jan 04 '17 at 14:31
  • There are many relevant posts, search this site for "score function". Some examples: http://stats.stackexchange.com/questions/95250/logistic-regression-maximum-likelihood-vs-misclassification/103893#103893 http://stats.stackexchange.com/questions/127042/why-isnt-logistic-regression-called-logistic-classification/127044#127044 http://stats.stackexchange.com/questions/104500/skills-hard-to-find-in-machine-learners/104533#104533 http://stats.stackexchange.com/questions/208529/when-is-it-appropriate-to-use-an-improper-scoring-rule/251654#251654 – kjetil b halvorsen Jan 04 '17 at 15:41

0 Answers0