Timeline for How should I approach this binary prediction problem?
Current License: CC BY-SA 4.0
19 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jun 24, 2019 at 11:02 | vote | accept | user1205901 - Слава Україні | ||
| Dec 9, 2018 at 10:29 | history | edited | user1205901 - Слава Україні | CC BY-SA 4.0 |
fixed typo
|
| Dec 26, 2016 at 21:51 | history | edited | user1205901 - Слава Україні | CC BY-SA 3.0 |
Clarified the sample size (eliminated the rounding)
|
| Dec 26, 2016 at 6:36 | comment | added | user1205901 - Слава Україні | @GeoMatt22 You mention the concept of a "tunable parameter", which wasn't one I was previously familiar with. I googled it, and also looked the webpage you linked in relation to the concept of a tunable parameter, i.e. this page about ROC. Can you explain a little further the link between ROCs and the concept of a tunable parameter, and the dataset I've described in my question? Is the basic point that I can look at different cutoffs (e.g. 3, 3.5, 4) and see which one causes me to get the best results in terms of the Peirce Skill Score? | |
| Dec 26, 2016 at 2:05 | history | edited | user1205901 - Слава Україні | CC BY-SA 3.0 |
Explain how I got Coefficient Alpha
|
| Dec 25, 2016 at 13:25 | history | tweeted | twitter.com/StackStats/status/813012792135524352 | ||
| Dec 24, 2016 at 9:39 | answer | added | Tavrock | timeline score: 1 | |
| Dec 24, 2016 at 8:58 | comment | added | GeoMatt22 | Note that your threshold is a tunable parameter, so the appropriate cutoff will depend on your evaluation criterion. As I was unfamiliar with your metric I Googled it, and actually the first hit may be relevant to you: A note on the maximum Peirce skill score (2007). | |
| Dec 24, 2016 at 7:10 | comment | added | user1205901 - Слава Україні | @rolando2 Thanks for the advice. I've rearranged things in my own data file so that now they are separated out. | |
| Dec 24, 2016 at 7:08 | comment | added | user1205901 - Слава Україні |
@Wayne The data ranges from the prediction of cancer with maximum confidence Cancer (4) to the prediction of no cancer with maximum confidence No Cancer (4). We can't say that No Cancer (3) and Cancer (2) are the same, but we could say there is a continuum, and the middle points in this continuum are Cancer (1) and No Cancer (1).
|
|
| Dec 24, 2016 at 5:48 | history | edited | user1205901 - Слава Україні | CC BY-SA 3.0 |
Implemented suggestions from answerer
|
| Dec 24, 2016 at 0:48 | comment | added | rolando2 | Re: your data structure, it's almost always better to have different variables (whether patient has cancer; how confident the assessment is) in different columns. Combining them as in "no cancer (3)" severely limits your options. | |
| Dec 23, 2016 at 17:20 | answer | added | Ricardo Cruz | timeline score: 2 | |
| Dec 23, 2016 at 17:10 | answer | added | GeoMatt22 | timeline score: 2 | |
| Dec 23, 2016 at 15:52 | comment | added | Wayne |
Can we say that No Cancer (3) is Cancer (2)? That would simplify your problem a bit.
|
|
| Dec 23, 2016 at 15:38 | answer | added | Jeremy Miles | timeline score: 7 | |
| Dec 23, 2016 at 15:27 | history | edited | Jeremy Miles |
edited tags
|
|
| Dec 23, 2016 at 11:39 | history | edited | user1205901 - Слава Україні | CC BY-SA 3.0 |
Fixed typo, added minor clarification
|
| Dec 23, 2016 at 3:23 | history | asked | user1205901 - Слава Україні | CC BY-SA 3.0 |