[Ensemble Learning] part 6 - 2

2023. 1. 20. 13:51
๐Ÿง‘๐Ÿป‍๐Ÿ’ป์šฉ์–ด ์ •๋ฆฌ

Accuracy
FPE
FNE
ROC curve

 

Supervised Learning์—์„œ model์˜ ์„ฑ๋Šฅ์€ ์–ด๋–ป๊ฒŒ ํ‰๊ฐ€ํ•  ์ˆ˜ ์žˆ์„๊นŒ์š”?

 

 

  • model์˜ Accuracy๋ฅผ ์ธก์ •ํ•ฉ๋‹ˆ๋‹ค.
    • ๊ฐ ๊ฒฝ์šฐ์— ๋Œ€ํ•ด ์˜ค์ฐจ๊ฐ€ ์–ผ๋งˆ๋‚˜ ์žˆ์—ˆ๋Š”์ง€ ํ‘œํ˜„ํ•˜๋Š” ๋ฐฉ๋ฒ•์€ Confusion matrix ์ž…๋‹ˆ๋‹ค.
  • False positive error:
    • ์‹ค์ œ๋กœ negative์ธ๋ฐ positive๋กœ ํŒ์ •ํ•œ ๊ฒฝ์šฐ์ž…๋‹ˆ๋‹ค.
  • False negative error:
    • ์‹ค์ œ๋กœ positive์ธ๋ฐ negative๋กœ ํŒ์ •ํ•œ ๊ฒฝ์šฐ์ž…๋‹ˆ๋‹ค.

 

๐Ÿ’ก Unbalanced Data set์˜ ๊ฒฝ์šฐ, Accuracy ์ด์™ธ์—๋„ precision๊ณผ recall ๊ฐ’์„ ๋™์‹œ์— ๋ณด์•„์•ผ model์˜ ์„ฑ๋Šฅ์„ ์ œ๋Œ€๋กœ ์ธก์ •ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

 

 

  • ROC Curve
    • ์„œ๋กœ ๋‹ค๋ฅธ Classifier์˜ ์„ฑ๋Šฅ์„ ์ธก์ •ํ•˜๋Š”๋ฐ ์‚ฌ์šฉํ•˜๋Š” curve์ž…๋‹ˆ๋‹ค.
    • ๊ฐ€๋กœ์ถ• FPR == 1 - TPR
    • ์„ธ๋กœ์ถ• sensitivity == recall

์ถœ์ฒ˜ : https://towardsdatascience.com/roc-curve-in-machine-learning-fea29b14d133

 

์„œ๋กœ ๋‹ค๋ฅธ classifier์— ๋Œ€ํ•ด์„œ ์—ฌ๋Ÿฌ ๊ฐœ์˜ ROC Curve๋ฅผ ๊ทธ๋ ธ์„ ๋•Œ, ์™ผ์ชฝ ์ƒ๋‹จ์œผ๋กœ ๊ฐˆ ์ˆ˜๋ก ์„ฑ๋Šฅ์ด ์ข‹์€ curve์ž„์„ ์˜๋ฏธํ•˜๊ฒŒ ๋ฉ๋‹ˆ๋‹ค.

 

 

  • Error measure
    • ๊ฒฝ์šฐ์— ๋”ฐ๋ผ recall or precision์ด ๋†’์•„์•ผ ํ•˜๋Š” ๊ฒฝ์šฐ๊ฐ€ ์กด์žฌํ•œ๋‹ค.

 

 

 

Supervised Learning์„ ์ด์šฉํ•œ App

 

์ตœ์‹  Machine Learning app

 

  • R-CNN
  • CNN - input feature ๋ฐ›์Œ / LSTM - output ๋ฌธ์žฅ ๊ตฌ์„ฑ
  • Semantic Segmentation
  • Pose estimation classification
  • Face detection
  • super resolution

 

 

problem

  • Large data samples
  • insufficient labels
  • Domain adaptation
  • transfer learning

 

 

 

BELATED ARTICLES

more