Probabilistic Prediction

Probabilistic Prediction

  • ๋ฏธ๋ž˜์— ์ผ์–ด๋‚  ์ผ์— ํ™•๋ฅ ์„ ์„ค์ •ํ•œ prediction, probabilistic forecasting์ด๋ผ๊ณ ๋„ ํ•จ

  • Example) prediction์ด "๋‚ด์ผ ๋น„๊ฐ€ ์˜จ๋‹ค"๋ผ๊ณ  ํ•œ๋‹ค๋ฉด probabilistic prediction์€ "๋‚ด์ผ ๋น„๊ฐ€ ์˜ฌ ํ™•๋ฅ ์ด 60%์ด๋‹ค"๋ผ๊ณ  ๋งํ•˜๋Š” ๊ฒƒ

Words of estimation probability (WEP)

  • WEP prediction: "๋‚ด์ผ ๋น„๊ฐ€ ์˜ฌ ๊ฒƒ ๊ฐ™๋‹ค"์ฒ˜๋Ÿผ ๋ถˆํ™•์‹ค์„ฑ์ด ๋‚ดํฌ๋œ ํ‘œํ˜„

  • WEP๋Š” ์ˆ˜์น˜๋กœ ๋ณ€ํ™˜ํ•˜๋Š” ๊ธฐ์ค€์ด ์—†์Œ

Odds

  • ์‚ฌ๊ฑด์˜ odds๋Š” ๋‹ค์Œ๊ณผ ๊ฐ™์ด ์ •์˜๋จ

O(E)=P(E)P(Ec)O(E) = {{P(E)} \over {P(E^c )}}
  • ๋ณดํ†ต E์™€ not E๋ฅผ ๋น„๊ตํ•จ

  • A์™€ B๋ฅผ ๋น„๊ตํ•  ์ˆ˜๋„ ์žˆ์Œ( = P(A) / P(B) )

  • Bayesian ๊ด€์ ์˜ odds: ์ฃผ์–ด์ง„ ๋ฐ์ดํ„ฐ D์— ๋Œ€ํ•œ hypothesis H์˜ odds

Bayes factors

P(M) = Marfan disease์— ๊ฑธ๋ ธ์„ ํ™•๋ฅ 

P(F) = ์ฆ์ƒ(features)์ด ๋‚˜ํƒ€๋‚  ํ™•๋ฅ 

  • posterior odds = Bayes factor * prior odds

  • Bayes factor๋ž€ likelihood์˜ ๋น„์œจ์„ ์˜๋ฏธํ•จ

  • Bayes factor๋Š” ๋ฐ์ดํ„ฐ์— ์˜ํ•ด ์ œ๊ณต๋œ 'evidence'์˜ ๊ฐ•๋„๋ฅผ ๋‚˜ํƒ€๋ƒ„

  • Bayes factor๊ฐ€ ํฌ๋”๋ผ๋„ prior odds๊ฐ€ ์ž‘์œผ๋ฉด odds๊ฐ€ ์ž‘์„ ์ˆ˜๋„ ์žˆ์Œ

Example) ์ „์ฒด ์ธ๊ตฌ์˜ 0.005์˜ ํ™•๋ฅ ๋กœ ์งˆ๋ณ‘์ด ๋ฐœ๋ณ‘ํ•œ๋‹ค. ์„ ๋ณ„ ๊ฒ€์‚ฌ ๊ฒฐ๊ณผ false positive๋Š” 0.05, false negative๋Š” 0.02์ด๋‹ค.

  1. ์งˆ๋ณ‘์˜ prior odds๋Š”?

likelihood table:

  1. ์ด ๋ฐ์ดํ„ฐ์—์„œ Bayes factor๋Š”?

    Bayes factor = ratio of likelihoods

    =P(T+โˆฃH+)P(T+โˆฃHโˆ’)=0.980.05=19.6= {{P(T_ + |H_ + )} \over {P(T_ + |H_ - )}} = {{0.98} \over {0.05}} = 19.6
  2. posterior odds๋Š”?

    = Bayes factor * prior odds = 19.6 * 0.00504 = 0.00985

  3. 1๊ณผ 2์˜ ๋‹ต๋ณ€์— ๊ทผ๊ฑฐํ•˜์—ฌ positive test(ํ•ด๋‹น ๊ฒ€์‚ฌ)๊ฐ€ ์ œ๊ณตํ•˜๋Š” evidence๊ฐ€ ๊ฐ•ํ•œ์ง€ ์•ฝํ•œ์ง€ ๊ตฌ๋ถ„ํ•  ์ˆ˜ ์žˆ์Œ?

    ๊ตฌ๋ถ„ ๊ฐ€๋Šฅํ•จ. bayes factor๊ฐ€ 19.6์œผ๋กœ ํ™˜์ž๊ฐ€ ๋ณ‘์— ๊ฑธ๋ ธ๋Š”์ง€์— ๋Œ€ํ•œ ๊ฐ•ํ•œ

    evidence๋ฅผ ๋‚˜ํƒ€๋ƒ„. ์•„๋ž˜์™€ ๊ฐ™์ด bayesian update table๋กœ๋„ ๋‚˜ํƒ€๋‚ผ ์ˆ˜ ์žˆ์Œ

Last updated