Archive

[๋ฒˆ์—ญ] Micro, Macro & Weighted Averages of F1 Score, Clearly Explained

geum 2022. 12. 20. 00:52

๐Ÿ’ฌ ์ตœ๋Œ€ํ•œ ๋งค๋„๋Ÿฝ๊ฒŒ ํ•ด์„ํ•˜๊ณ ์ž ๋…ธ๋ ฅํ–ˆ์ง€๋งŒ ์–ด์ƒ‰ํ•œ ๋ฌธ์žฅ์ด ์žˆ์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ํ”ผ๋“œ๋ฐฑ์€ ์–ธ์ œ๋‚˜ ํ™˜์˜์ž…๋‹ˆ๋‹ค ๐Ÿ™‚


์›๋ณธ ๊ธ€ ์ฃผ์†Œ: https://towardsdatascience.com/micro-macro-weighted-averages-of-f1-score-clearly-explained-b603420b292f

 

Micro, Macro & Weighted Averages of F1 Score, Clearly Explained

Understanding the concepts behind the micro average, macro average and weighted average of F1 score in multi-class classification

towardsdatascience.com

 

F1 Score(F-measure๋ผ๊ณ ๋„ ํ•จ)๋Š” ๋ถ„๋ฅ˜ ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ ํ‰๊ฐ€์— ๋งŽ์ด ์‚ฌ์šฉ๋˜๋Š” ํ‰๊ฐ€ ์ง€ํ‘œ์ž…๋‹ˆ๋‹ค. ๋‹ค์ค‘ ํด๋ž˜์Šค ๋ถ„๋ฅ˜์˜ ๊ฒฝ์šฐ, F1 score ๊ณ„์‚ฐ์„ ์œ„ํ•ด ์„œ๋กœ ๋‹ค๋ฅธ ํ‰๊ท ํ™” ๋ฐฉ๋ฒ•(macro, weighted, micro)์„ ์‚ฌ์šฉํ•˜์—ฌ ๋ถ„๋ฅ˜ ๊ฒฐ๊ณผ๋ฅผ ํ™•์ธํ•ฉ๋‹ˆ๋‹ค. ์ด ๊ธ€์—์„œ๋Š” ๊ฐ ํ‰๊ท ์˜ ์˜๋ฏธ์™€ ๊ณ„์‚ฐ ๋ฐฉ๋ฒ•, ๊ฒฐ๊ณผ ๋ถ„์„์„ ์œ„ํ•ด ๋ฌด์—‡์„ ๊ณจ๋ผ์•ผ ํ•˜๋Š”์ง€์— ๋Œ€ํ•ด ์•Œ์•„๋ด…๋‹ˆ๋‹ค.

 

๋ณต์Šต(Optional)

* Layman definition: ๋น„์ „๋ฌธ๊ฐ€์ ์ธ ์ •์˜; ์‰ฌ์šด ์ •์˜

 

Precision

Layman definition: ์˜ˆ์ธก ๊ฒฐ๊ณผ๊ฐ€ positive์ธ ๊ฒƒ ์ค‘ ์‹ค์ œ positive์˜ ์ˆ˜

 

๊ณ„์‚ฐ:

 

Recall

Layman definition: ์‹ค์ œ positive์ธ ๊ฒƒ ์ค‘ ์˜ˆ์ธก ๊ฒฐ๊ณผ๋„ positive์ธ ๊ฒƒ์˜ ์ˆ˜

 

๊ณ„์‚ฐ:

Precision๊ณผ recall์˜ ์ˆ˜์‹์€ ๊ต‰์žฅํžˆ ์œ ์‚ฌํ•ด ๋ณด์ž…๋‹ˆ๋‹ค. ์œ ์ผํ•œ ์ฐจ์ด์ ์€ ๋ถ„๋ชจ์˜ ๋‘ ๋ฒˆ์งธ ํ•ญ์ด precision์—์„œ๋Š” False Positive์ด๊ณ  recall์—์„œ๋Š” False Negative๋ผ๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.

 

F1 Score

๋ชจ๋ธ ์„ฑ๋Šฅ์„ ์ข…ํ•ฉ์ ์œผ๋กœ ํ‰๊ฐ€ํ•˜๋ ค๋ฉด precision๊ณผ recall์„ ๋ชจ๋‘ ํ™•์ธํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. F1 Score๋Š” ๋‘ ๊ฐ’์„ ๋ชจ๋‘ ๊ณ ๋ คํ•˜๋Š” ์œ ์šฉํ•œ ํ‰๊ฐ€ ์ง€ํ‘œ ์—ญํ• ์„ ํ•ฉ๋‹ˆ๋‹ค.

 

์ •์˜: ๋ชจ๋ธ ์„ฑ๋Šฅ์˜ ๊ท ํ˜• ์žกํžŒ ์š”์•ฝ์„ ์œ„ํ•œ precision๊ณผ recall์˜ ์กฐํ™” ํ‰๊ท 

 

๊ณ„์‚ฐ:

 

TP, FP, FN์œผ๋กœ F1 Score๋ฅผ ํ‘œํ˜„ํ•œ๋‹ค๋ฉด ๋‹ค์Œ๊ณผ ๊ฐ™์€ ์‹์„ ์–ป์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

 

์˜ˆ์‹œ ์„ค์ •

ํ‰๊ท  F1 Score์˜ ๊ฐœ๋…์„ ์ดํ•ดํ•˜๊ธฐ ์œ„ํ•ด ์ด ํŠœํ† ๋ฆฌ์–ผ์—์„œ๋Š” ์•„๋ž˜ ์˜ˆ์‹œ๋ฅผ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค.

 

๋น„ํ–‰๊ธฐ, ๋ฐฐ, ์ž๋™์ฐจ 3๊ฐœ์˜ ํด๋ž˜์Šค๋ฅผ ๊ฐ€์ง„ ๋ฉ€ํ‹ฐ ํด๋ž˜์Šค ๋ฐ์ดํ„ฐ์…‹์œผ๋กœ ์ด๋ฏธ์ง€ ๋ถ„๋ฅ˜ ๋ชจ๋ธ์„ ํ•™์Šต์‹œํ‚จ๋‹ค๊ณ  ์ƒ์ƒํ•ด๋ณด์„ธ์š”. 10๊ฐœ์˜ ํ…Œ์ŠคํŠธ ์ด๋ฏธ์ง€ ํด๋ž˜์Šค๋ฅผ ๋ถ„๋ฅ˜ํ•˜๊ธฐ ์œ„ํ•ด ์ด ๋ชจ๋ธ์„ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค.

 

์ถœ์ฒ˜: https://towardsdatascience.com/micro-macro-weighted-averages-of-f1-score-clearly-explained-b603420b292f

 

sklearn.metrics.classfication_report ๋ช…๋ น์–ด๋ฅผ ์‹คํ–‰ํ•˜๋ฉด ์•„๋ž˜์˜ ๋ถ„๋ฅ˜ ๋ฆฌํฌํŠธ๋ฅผ ์–ป์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

 

์ถœ์ฒ˜: https://towardsdatascience.com/micro-macro-weighted-averages-of-f1-score-clearly-explained-b603420b292f

  

ํด๋ž˜์Šค๋ณ„ ์ ์ˆ˜(์ฃผํ™ฉ์ƒ‰ ์ปฌ๋Ÿผ)์™€ ํ‰๊ท  ์ ์ˆ˜๋Š” ์šฐ๋ฆฌ๊ฐ€ ๋‚ด๋ฆฌ๊ฒŒ ๋  ๊ฒฐ์ •์˜ ์ดˆ์ ์ž…๋‹ˆ๋‹ค.

 

10๊ฐœ์˜ ํ…Œ์ŠคํŠธ ์ด๋ฏธ์ง€ ๋ชฉ๋ก์„ ๋ณด๋ฉด ๋ฐ์ดํ„ฐ์…‹์ด ๋ถˆ๊ท ํ˜•ํ•œ ๊ฒƒ์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.(10๊ฐœ ์ด๋ฏธ์ง€ ์ค‘ ๋‹จ ํ•˜๋‚˜๋งŒ์ด '๋ฐฐ' ํด๋ž˜์Šค์— ํ•ด๋‹น) ๋”ฐ๋ผ์„œ ์˜ˆ์ธก๊ณผ ์ •๋‹ต์˜ ์ผ์น˜ ๋น„์œจ(์ •ํ™•๋„)์€ ๋ชจ๋ธ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•˜๋Š” ๋ฐ ํšจ๊ณผ์ ์ธ ๋ฐฉ๋ฒ•์ด ์•„๋‹™๋‹ˆ๋‹ค.

 

๋Œ€์‹  ๋ชจ๋ธ ์˜ˆ์ธก์— ๋Œ€ํ•œ ์ „์ฒด์ ์ธ ์ดํ•ด๋ฅผ ์œ„ํ•ด confusion matrix๋ฅผ ์‚ดํŽด๋ด…์‹œ๋‹ค.

 

์ถœ์ฒ˜: https://towardsdatascience.com/micro-macro-weighted-averages-of-f1-score-clearly-explained-b603420b292f

 

์œ„์˜ confusion matrix๋ฅผ ์ด์šฉํ•˜๋ฉด ์•„๋ž˜์™€ ๊ฐ™์ด True Positive(TP), False Positive(FP) ๋ฐ False Negative(FN)์˜ ์ž„๊ณ„๊ฐ’์„ ๊ตฌํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

 

์ถœ์ฒ˜: https://towardsdatascience.com/micro-macro-weighted-averages-of-f1-score-clearly-explained-b603420b292f

 

์œ„ ํ‘œ๋Š” 3๊ฐœ์˜ ํด๋ž˜์Šค ๊ฐ๊ฐ์— ๋Œ€ํ•œ preicision, recall ๊ทธ๋ฆฌ๊ณ  F1 Score์˜ ํด๋ž˜์Šค๋ณ„ ๊ฐ’์„ ์ž˜ ๊ณ„์‚ฐํ•  ์ˆ˜ ์žˆ๊ฒŒ ํ•ฉ๋‹ˆ๋‹ค.

 

๋‹ค์ค‘ ๋ถ„๋ฅ˜์—์„œ๋Š” ์ด์ง„ ๋ถ„๋ฅ˜์ฒ˜๋Ÿผ ์ „์ฒด์— ๋Œ€ํ•œ ํ•˜๋‚˜์˜ F1 Score ๋Œ€์‹  OvR(One-vs-Rest) ์ ‘๊ทผ ๋ฐฉ์‹์œผ๋กœ ๊ฐ ํด๋ž˜์Šค์— ๋Œ€ํ•œ F1 Score๋ฅผ ๊ณ„์‚ฐํ•œ๋‹ค๋Š” ๊ฒƒ์ด ์ค‘์š”ํ•œ ์ ์ž…๋‹ˆ๋‹ค.

 

OvR ๋ฐฉ์‹์—์„œ๋Š” ๊ฐ ํด๋ž˜์Šค์— ๋‹ค๋ฅธ ๋ถ„๋ฅ˜๊ธฐ๊ฐ€ ์žˆ๋Š” ๊ฒƒ์ฒ˜๋Ÿผ ์„ฑ๋Šฅ์„ ๊ฐœ๋ณ„์ ์œผ๋กœ ๊ฒฐ์ •ํ•ฉ๋‹ˆ๋‹ค. ํด๋ž˜์Šค๋ณ„ ์„ฑ๋Šฅ์€ ๋‹ค์Œ๊ณผ ๊ฐ™์Šต๋‹ˆ๋‹ค.

 

์ถœ์ฒ˜: https://towardsdatascience.com/micro-macro-weighted-averages-of-f1-score-clearly-explained-b603420b292f

 

ํ•˜์ง€๋งŒ ํด๋ž˜์Šค๋ณ„๋กœ ์—ฌ๋Ÿฌ ๊ฐœ์˜ F1 Score๋ฅผ ๊ฐ–๋Š” ๋Œ€์‹ , ์ „์ฒด ์„ฑ๋Šฅ์„ ๋‚˜ํƒ€๋‚ด๋Š” ํ•˜๋‚˜์˜ ์ˆซ์ž๋ฅผ ์–ป๊ธฐ ์œ„ํ•ด ํ‰๊ท ์„ ์ทจํ•˜๋Š” ๊ฒƒ์ด ๋” ๋‚˜์„ ๊ฒƒ ๊ฐ™์Šต๋‹ˆ๋‹ค.

 

์ด์ œ ๋ถ„๋ฅ˜ ๋ฆฌํฌํŠธ์˜ ์„ธ ๊ฐ€์ง€ ํ‰๊ท  F1 Score๋กœ ์ด์–ด์ง„ ํ‰๊ท ํ™” ๋ฐฉ๋ฒ•์— ๋Œ€ํ•ด ์•Œ์•„๋ณด๊ฒ ์Šต๋‹ˆ๋‹ค.

 

Macro Average

Macro Averaging์€ ์•„๋งˆ ์ˆ˜๋งŽ์€ ํ‰๊ท ํ™” ๋ฐฉ๋ฒ• ์ค‘ ๊ฐ€์žฅ ๊ฐ„๋‹จ ๋ฐฉ๋ฒ•์ผ ๊ฒƒ์ž…๋‹ˆ๋‹ค.

 

Macro-averaged F1 Score(๋˜๋Š” macro F1 Score)๋Š” ๋ชจ๋“  ํด๋ž˜์Šค์˜ ์‚ฐ์ˆ  ํ‰๊ท (๊ฐ€์ค‘์น˜๋ฅผ ๋‘์ง€ ์•Š๋Š” ํ‰๊ท )์„ ์ด์šฉํ•ด ๊ณ„์‚ฐ๋ฉ๋‹ˆ๋‹ค.

 

์ด ๋ฐฉ๋ฒ•์€ ๊ฐ ํด๋ž˜์Šค์˜ support value์— ๊ด€๊ณ„์—†์ด ๋ชจ๋“  ํด๋ž˜์Šค๋ฅผ ๋™๋“ฑํ•˜๊ฒŒ ์ฒ˜๋ฆฌํ•ฉ๋‹ˆ๋‹ค.

 

์ถœ์ฒ˜: https://towardsdatascience.com/micro-macro-weighted-averages-of-f1-score-clearly-explained-b603420b292f

 

์œ„์—์„œ ๊ณ„์‚ฐํ•œ ๊ฐ’ 0.58์€ ๋ถ„๋ฅ˜ ๋ฆฌํฌํŠธ์—์„œ์˜ macro avg์™€ ๋™์ผํ•ฉ๋‹ˆ๋‹ค.

 

์ถœ์ฒ˜: https://towardsdatascience.com/micro-macro-weighted-averages-of-f1-score-clearly-explained-b603420b292f

 

Weighted Average

Weighted-averaged F1 Score๋Š” ๊ฐ ํด๋ž˜์Šค์˜ support๋ฅผ ๊ณ ๋ คํ•˜์—ฌ ๋ชจ๋“  ํด๋ž˜์Šค์˜ F1 Score ํ‰๊ท ์„ ๊ตฌํ•ฉ๋‹ˆ๋‹ค.

 

Support๋Š” ๋ฐ์ดํ„ฐ์…‹์—์„œ ํด๋ž˜์Šค์˜ ์‹ค์ œ ๋ฐœ์ƒ ํšŸ์ˆ˜๋ฅผ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, '๋ฐฐ'์˜ support value๊ฐ€ 1์ด๋ผ๋ฉด '๋ฐฐ' ๋ ˆ์ด๋ธ”์ด ์žˆ๋Š” ๊ด€์ธก์น˜๊ฐ€ ํ•˜๋‚˜์ž„์„ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค.

 

'Weight'๋Š” ์ „์ฒด support value์— ๋Œ€ํ•œ ํด๋ž˜์Šค์˜ ์ƒ๋Œ€์ ์ธ support ๋น„์œจ์„ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค.

 

์ถœ์ฒ˜: https://towardsdatascience.com/micro-macro-weighted-averages-of-f1-score-clearly-explained-b603420b292f

 

Weighted averaging์˜ ๊ฒฝ์šฐ, ์ถœ๋ ฅ๋œ ํ‰๊ท ์€ ์ฃผ์–ด์ง„ ํด๋ž˜์Šค์— ํ•ด๋‹นํ•˜๋Š” ๋ฐ์ดํ„ฐ ์ˆ˜์— ๋”ฐ๋ผ ๊ฐ€์ค‘์น˜๊ฐ€ ๋ถ€์—ฌ๋œ ๊ฐ ํด๋ž˜์Šค์˜ ๊ธฐ์—ฌ๋„์— ๋Œ€ํ•ด ์„ค๋ช…ํ•˜๊ณ  ์žˆ์„ ๊ฒƒ์ž…๋‹ˆ๋‹ค.

 

์œ„์—์„œ ๊ณ„์‚ฐํ•œ ๊ฐ’ 0.64๋Š” ๋ถ„๋ฅ˜ ๋ฆฌํฌํŠธ์—์„œ์˜ weighted avg์™€ ๋™์ผํ•ฉ๋‹ˆ๋‹ค.

 

Micro Average

Micro Averaging์€ TP, FN, FP ๊ฐ๊ฐ์˜ ์ดํ•ฉ์„ ๊ตฌํ•ด global average F1 Score๋ฅผ ๊ณ„์‚ฐํ•ฉ๋‹ˆ๋‹ค.

 

๋จผ์ € ๋ชจ๋“  ํด๋ž˜์Šค์— ๋Œ€ํ•ด ์ƒ๋Œ€์ ์ธ TP, FP, FN์„ ๊ตฌํ•˜๊ณ  micro F1 Score๋ฅผ ์–ป๊ธฐ ์œ„ํ•ด F1 ๊ณต์‹์— ๊ฐ’์„ ๋Œ€์ž…ํ•ฉ๋‹ˆ๋‹ค.

 

์ถœ์ฒ˜: https://towardsdatascience.com/micro-macro-weighted-averages-of-f1-score-clearly-explained-b603420b292f

 

๋ถ„๋ฅ˜ ๋ฆฌํฌํŠธ์—์„œ micro F1 Score ๊ฐ’์ธ 0.6์ด ์™œ micro avg๊ฐ€ ์•„๋‹Œ accuracy๋กœ ํ‘œ๊ธฐ๋˜์–ด ์žˆ๋Š”์ง€ ๊ถ๊ธˆํ•  ๊ฒƒ์ž…๋‹ˆ๋‹ค.

 

์ถœ์ฒ˜: https://towardsdatascience.com/micro-macro-weighted-averages-of-f1-score-clearly-explained-b603420b292f

 

์ด๋Ÿฌํ•œ ์ด์œ ๋Š” micro averaging์ด ๋ชจ๋“  ๊ด€์ธก์น˜ ์ค‘์—์„œ ์ •ํ™•ํ•˜๊ฒŒ ๋ถ„๋ฅ˜ํ•œ ๊ฒƒ์˜ ๋น„์œจ์„ ๊ณ„์‚ฐํ•˜๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. ์ด ์ ์„ ์ƒ๊ฐํ•œ๋‹ค๋ฉด micro averaging์˜ ์ •์˜๋Š” ์ „์ฒด accuracy ๊ณ„์‚ฐ์„ ์œ„ํ•ด ์‚ฌ์šฉํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.

 

๋˜ํ•œ, precision๊ณผ recall์— micro averaging์„ ์ ์šฉํ•ด๋„ 0.6์ด๋ผ๋Š” ๊ฐ™์€ ๊ฐ’์„ ์–ป๊ฒŒ ๋ฉ๋‹ˆ๋‹ค.

 

์ถœ์ฒ˜: https://towardsdatascience.com/micro-macro-weighted-averages-of-f1-score-clearly-explained-b603420b292f

 

์ด ๊ฒฐ๊ณผ๋Š” ๊ฐ๊ฐ์˜ ๊ด€์ธก์น˜๊ฐ€ ํ•˜๋‚˜์˜ ๋ ˆ์ด๋ธ”(๋น„ํ–‰๊ธฐ or ๋ฐฐ or ์ฐจ)๋งŒ์„ ๊ฐ€์ง€๋Š” ๋‹ค์ค‘ ๋ถ„๋ฅ˜ ๋ฌธ์ œ์—์„œ micro-F1, micro-precision, micro-recall๊ณผ accuracy๋Š” ๋ชจ๋‘ ๊ฐ™์€ ๊ฐ’์„ ๊ฐ€์ง„๋‹ค๋Š” ๊ฒƒ์„ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค. (์ด ์˜ˆ์ œ์—์„œ๋Š” 0.6)

 

๊ทธ๋ฆฌ๊ณ  ์ด๋Š” micro-F1, micro-precision, micro-recall์ด ๋™์ผํ•œ ๊ฐ’์ด๊ธฐ ๋•Œ๋ฌธ์— ๋ถ„๋ฅ˜ ๋ฆฌํฌํŠธ๋„ ํ•˜๋‚˜์˜ accuracy ๊ฐ’๋งŒ ํ‘œ์‹œํ•˜๋ฉด ๋˜๋Š” ์ด์œ ์ž…๋‹ˆ๋‹ค. 

 

์–ด๋–ค average ๋ฐฉ๋ฒ•์„ ๊ณจ๋ผ์•ผ ํ• ๊นŒ?

๋ชจ๋“  ํด๋ž˜์Šค๊ฐ€ ๋˜‘๊ฐ™์ด ์ค‘์š”ํ•œ ๋ถˆ๊ท ํ˜• ๋ฐ์ดํ„ฐ์…‹์œผ๋กœ ์ž‘์—…ํ•˜๊ฒŒ ๋œ๋‹ค๋ฉด ๋ชจ๋“  ํด๋ž˜์Šค๋ฅผ ๋™๋“ฑํ•˜๊ฒŒ ์ฒ˜๋ฆฌํ•˜๋Š” macro average๊ฐ€ ์ข‹์€ ์„ ํƒ์ง€๊ฐ€ ๋  ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด๊ฒƒ์€ ๋น„ํ–‰๊ธฐ, ๋ฐฐ, ์ž๋™์ฐจ ๋ถ„๋ฅ˜์™€ ๊ด€๋ จ๋œ ์˜ˆ์ œ์ผ ๊ฒฝ์šฐ macro-F1 Score๋ฅผ ์‚ฌ์šฉํ•œ๋‹ค๋Š” ๊ฒƒ์ด๋ผ๋Š” ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค.

 

๋งŒ์•ฝ ๋ถˆ๊ท ํ˜•ํ•œ ๋ฐ์ดํ„ฐ์…‹์ด๋ฉด์„œ ๊ฐœ์ˆ˜๊ฐ€ ๋” ๋งŽ์€ ํด๋ž˜์Šค์— ๋†’์€ ๊ธฐ์—ฌ๋„๋ฅผ ์ฃผ๊ณ  ์‹ถ๋‹ค๋ฉด, weighted average๊ฐ€ ์„ ํ˜ธ๋ฉ๋‹ˆ๋‹ค. Weighted averaging์—์„œ๋Š” ํด๋ž˜์Šค๋ณ„ ๊ฐœ์ˆ˜์— ๋”ฐ๋ผ  F1 ํ‰๊ท ์— ๋Œ€ํ•œ ๊ฐ ํด๋ž˜์Šค์˜ ๊ธฐ์—ฌ๋„์— ๊ฐ€์ค‘์น˜๊ฐ€ ๋ถ€์—ฌ๋˜๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค.

 

๋งŒ์•ฝ ๋ฐ์ดํ„ฐ์…‹์˜ ๊ท ํ˜•์ด ์ž˜ ๋งž๊ณ  ํด๋ž˜์Šค์— ์ƒ๊ด€์—†์ด ์ „์ฒด ์„ฑ๋Šฅ์„ ์‰ฝ๊ฒŒ ์ดํ•ดํ•˜๊ณ ์ž ํ•œ๋‹ค๊ณ  ๊ฐ€์ •ํ•ด๋ด…์‹œ๋‹ค. ์ด๋Ÿฐ ๊ฒฝ์šฐ์—๋Š” micro F1 Score์ธ accuracy๋ฅผ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.