โ€œํ™•๋ฅ ๊ณผ ํ†ต๊ณ„(MATH230)โ€ ์ˆ˜์—…์—์„œ ๋ฐฐ์šด ๊ฒƒ๊ณผ ๊ณต๋ถ€ํ•œ ๊ฒƒ์„ ์ •๋ฆฌํ•œ ํฌ์ŠคํŠธ์ž…๋‹ˆ๋‹ค. ์ „์ฒด ํฌ์ŠคํŠธ๋Š” Probability and Statistics์—์„œ ํ™•์ธํ•˜์‹ค ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค ๐ŸŽฒ

11 minute read

โ€œํ™•๋ฅ ๊ณผ ํ†ต๊ณ„(MATH230)โ€ ์ˆ˜์—…์—์„œ ๋ฐฐ์šด ๊ฒƒ๊ณผ ๊ณต๋ถ€ํ•œ ๊ฒƒ์„ ์ •๋ฆฌํ•œ ํฌ์ŠคํŠธ์ž…๋‹ˆ๋‹ค. ์ „์ฒด ํฌ์ŠคํŠธ๋Š” Probability and Statistics์—์„œ ํ™•์ธํ•˜์‹ค ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค ๐ŸŽฒ

Poisson Distribution

<ํ‘ธ์•„์†ก ๋ถ„ํฌ Poisson Distribution>๋Š” ์ดํ•ญ ๋ถ„ํฌ $\text{BIN}(n, p)$์˜ ํŠน์ˆ˜ํ•œ ๊ฒฝ์šฐ์ด๋‹ค. $\text{BIN}(n, p)$์—์„œ $n$์ด ๋ฌดํ•œ๋Œ€๋กœ ์ปค์ง€๊ณ , $p$๊ฐ€ ์•„์ฃผ์•„์ฃผ ์ž‘์•„์งˆ ๋•Œ, ๋ถ„ํฌ๋Š” ํ‘ธ์•„์†ก ๋ถ„ํฌ๋ฅผ ๋งŒ์กฑํ•˜๊ฒŒ ๋œ๋‹ค!

๊ทธ๋ ‡๋‹ค๋ฉด ๋ณธ๋ž˜ BIN์ด๋˜๊ฑธ ์™œ ํ‘ธ์•„์†ก ๋ถ„ํฌ๋กœ ํ•ด์„ํ•˜๋Š” ๊ฑธ๊นŒ? ์ด ์งˆ๋ฌธ์— ๋Œ€ํ•œ ๋‹ต์€ ์•„๋ž˜์˜ ์œ ํŠœ๋ธŒ ์˜์ƒ์—์„œ ์ •๋ง ์ž˜ ์„ค๋ช…ํ•˜๊ณ  ์žˆ๋‹ค. ํ•œ๋ฒˆ ๋ณด๊ณ  ์˜ค์ž.

๐Ÿ‘‰ Youtube - ํ‘ธ์•„์†ก๋ถ„ํฌ ์†Œ๊ฐœ

์ฆ‰, $n$๊ณผ $p$์˜ ๊ฐ’์„ ๋‹ค๋ฃฐ ์ˆ˜๋„ ์—†๊ณ  ์ •์˜ํ•  ์ˆ˜๋„ ์—†์„ ๋•Œ, ํ‘ธ์•„์†ก์€ $np$๋ฅผ $\lambda$๋กœ ๋‘๊ณ  ์ƒˆ๋กœ์šด ํ˜•ํƒœ์˜ ๋ถ„ํฌ๋ฅผ ์œ ๋„ํ•œ ๊ฒƒ์ด๋‹ค. ๋˜๋Š” ํ‰๊ท ๊ฐ’์ธ $\lambda$๋ฅผ ์•„๋Š” ์ƒํƒœ์—์„œ ์œ ๋„ํ•œ ๋ถ„ํฌ๋ผ๊ณ  ๋ณผ ์ˆ˜ ์žˆ์„ ๊ฒƒ ๊ฐ™๋‹ค.

Definition. Poisson Distribution

A Poisson random variable $X$ with parameter $\lambda > 0$, denoted as $X \sim \text{POI}(\lambda)$, and it has a pmf $f(x)$ as

\[f(x) = e^{-\lambda} \frac{\lambda^x}{x!} \quad \text{for} \quad x=0, 1, \dots\]

์ด ํ‘ธ์•„์†ก ๋ถ„ํฌ๊ฐ€ ์ •๋ง pmf์ธ์ง€ ๊ฒ€์ฆํ•ด๋ณด์ž. ํ™•๋ฅ ์˜ ๅˆ์ด 1์ด ๋จ์„ ๋ณด์ด๋ฉด ๋œ๋‹ค.

\[\begin{aligned} \sum f(x) &= e^{-\lambda} \sum_{x=0} \frac{\lambda^x}{x!} \\ &= e^{-\lambda} e^\lambda = 1 \end{aligned}\]

์•ž์—์„œ ํ‘ธ์•„์†ก ๋ถ„ํฌ๋Š” ์ดํ•ญ๋ถ„ํฌ์˜ ํŠน์ˆ˜ํ•œ ๊ฒฝ์šฐ๋ผ๊ณ  ์†Œ๊ฐœํ–ˆ๋‹ค. ์ด๊ฒƒ์„ ํ™•์ธํ•ด๋ณด์ž.

Derivation.

Let $X \sim \text{BIN}(n, p) = \text{BIN}(n, \frac{\lambda}{n})$, then pmf $f_n (x)$ is

\[f_n (x) = \binom{n}{x} p^x (1-p)^{n-x} = \binom{n}{x} \left( \frac{\lambda}{n}\right)^x \left( 1 - \frac{\lambda}{n}\right)^{n-x}\]

์œ„์˜ ์‹์—์„œ $\binom{n}{x}$๋ฅผ ํ’€์–ด์„œ ์จ๋ณด๋ฉด ์•„๋ž˜์™€ ๊ฐ™๊ณ , ์ด๊ฒƒ์„ ์ž˜ ์ •๋ฆฌํ•ด๋ณด์ž.

\[\begin{aligned} f_n (x) &= \binom{n}{x} \left( \frac{\lambda}{n}\right)^x \left( 1 - \frac{\lambda}{n}\right)^{n-x} \\ &= \frac{n!}{x!(n-x)!} \frac{\lambda^x}{n^x} \left( 1 - \frac{\lambda}{n}\right)^n \left( 1 - \frac{\lambda}{n}\right)^{-x} \\ &= \frac{\lambda^x}{x!} \cdot \left( 1 - \frac{\lambda}{n}\right)^n \cdot \frac{n!}{(n-x)!} \left(\frac{1}{n}\right)^x \left( 1 - \frac{\lambda}{n}\right)^{-x} \end{aligned}\]

์ด์ œ ์œ„์˜ ์‹์—์„œ $n \rightarrow \infty$๋ฅผ ์ทจํ•˜์ž!

\[\begin{aligned} \lim_{n \rightarrow \infty} f_n (x) &= \lim_{n \rightarrow \infty} \frac{\lambda^x}{x!} \cdot \left( 1 - \frac{\lambda}{n}\right)^n \cdot \frac{n!}{(n-x)!} \left(\frac{1}{n}\right)^x \left( 1 - \frac{\lambda}{n}\right)^{-x} \\ &= \lim_{n \rightarrow \infty} \frac{\lambda^x}{x!} \cdot e^{-\lambda} \cdot \frac{n(n-1)\cdots(n-x+1)}{n^x} \cdot \frac{(n-\lambda)^{-x}}{n^{-x}} \\ &= \frac{\lambda^x}{x!} \cdot e^{-\lambda} \cdot 1 \cdot 1 \\ &= \frac{\lambda^x}{x!} e^{-\lambda} \end{aligned}\]

$\blacksquare$

์œ„์˜ ์œ ๋„ ๊ณผ์ •์—์„œ๋Š” ์ดํ•ญ ๋ถ„ํฌ๋ฅผ ์‚ฌ์šฉํ–ˆ์ง€๋งŒ, ๋ฏธ๋ถ„๋ฐฉ์ •์‹์œผ๋กœ๋„ ํ‘ธ์•„์†ก ๋ถ„ํฌ๋ฅผ ์œ ๋„ํ•  ์ˆ˜ ์žˆ๋‹ค๊ณ  ํ•œ๋‹ค. ์œ ๋„ ๊ณผ์ •์— ๋Œ€ํ•œ ์˜์ƒ์„ ๋งํฌ๋กœ ๊ฑธ์–ด๋‘”๋‹ค. ๐Ÿ‘‰ YouTube - ํ‘ธ์•„์†ก ๋ถ„ํฌ, ๋ฏธ๋ถ„๋ฐฉ์ •์‹์œผ๋กœ ์œ ๋„

Theorem. Law of Rare Events

$n$์ด ๋ฌดํ•œํžˆ ์ปค์ง€๊ฒŒ ๋˜๋ฉด, ์ž์—ฐ์Šค๋Ÿฝ๊ฒŒ ํ™•๋ฅ  $p=\dfrac{\lambda}{n}$๋Š” ์ž‘์•„์ง€๊ฒŒ ๋œ๋‹ค. ํ•˜์ง€๋งŒ, ์ดํ•ญ ๋ถ„ํฌ์˜ ์„ฑ์งˆ์— ๋”ฐ๋ผ ์—ฌ์ „ํžˆ ํ‰๊ท ๊ณผ ๋ถ„์‚ฐ์€ ์•„๋ž˜์™€ ๊ฐ™์„ ๊ฒƒ์ด๋‹ค.

  • $\displaystyle E[X] = \lim_{n\rightarrow\infty} np = \lambda$
  • $\displaystyle \text{Var}(X) = \lim_{n\rightarrow\infty} n \frac{\lambda}{n} \left( 1 - \frac{\lambda}{n}\right) = \lambda$

์ด๋Ÿฐ ์ƒํ™ฉ์— ๋Œ€ํ•ด ๊ธฐ์ˆ ํ•œ ์ •๋ฆฌ๊ฐ€ ๋ฐ”๋กœ <Law of rare event>์ด๋‹ค ๐Ÿ˜Ž

์œ„์˜ ๋ช…์ œ์— ๋Œ€ํ•œ ์ฆ๋ช…์€ ํ‰๊ท ๊ณผ ๋ถ„์‚ฐ์˜ ์ •์˜์— ์ž…๊ฐํ•ด ์‹์„ ์ „๊ฐœํ•˜๋ฉด ๋œ๋‹ค. ์ฆ๋ช…์€ ์ถ”ํ›„์— ๊ธฐ์ˆ ํ•˜๊ฒ ๋‹ค.


Bernoulli Process & Poisson Process

Bernoulli Process

<Poission Process>๋ฅผ ๋‹ค๋ฃจ๊ธฐ ์œ„ํ•ด์„  ๋จผ์ € <Bernoulli Process>์— ๋Œ€ํ•ด ์•Œ์•„์•ผ ํ•œ๋‹ค.

Definition. Bernoulli Process

The <Bernoulli process> is a sequence of independent Bernoulli trials.

At each trial $X_i$,

  • $P(H) = P(X_i = 1) = p$
  • $P(T) = P(X_i = 0) = 1-p$

์ฆ‰, ๋ฒ ๋ฅด๋ˆ„์ด ์‹œํ–‰์€ Bernoulli RV Sequence $X = \{ X_n : n=1, 2, \dots \}$๋ผ๊ณ  ๋ณผ ์ˆ˜ ์žˆ๋‹ค.

\[X_i \sim \text{Ber}(p) \quad \text{and} \quad X \sim \text{BP}(p)\]

์ด๋Ÿฐ ๋ฒ ๋ฅด๋ˆ„์ด ํ”„๋กœ์„ธ์Šค์˜ ์˜ˆ๋กœ๋Š”

  • ๋งค์ผ ์ฝ”์Šคํ”ผ ์ง€์ˆ˜์˜ ์ƒ์Šน/ํ•˜๋ฝ์— ๋Œ€ํ•œ binary sequence
  • ์ฃผ์–ด์ง„ time interval์— ์‹ ํ˜ธ๊ฐ€ ์ˆ˜์‹ ๋˜๋Š”์ง€ ์•„๋‹Œ์ง€์— ๋Œ€ํ•œ binary seq.

Poisson Process

์ด๋ฒˆ์—๋Š” BP์—์„œ ๊ทนํ•œ์„ ์ทจํ•ด time intervel์˜ ๊ฐ„๊ฒฉ์„ ์•„์ฃผ์•„์ฃผ ์ค„์ธ, ๊ทธ๋ž˜์„œ ๊ฒฐ๊ตญ continuousํ•œ ์‹œ๊ฐ„์ถ• ์œ„์—์„œ ์‹œํ–‰๋˜๋Š” <Poisson Process>์— ๋Œ€ํ•ด ์‚ดํŽด๋ณด์ž. ์•„๋ž˜์— ๊ธฐ์ˆ ๋˜๋Š” ๋‚ด์šฉ์€ ์•„๋ž˜์˜ ์œ ํŠœ๋ธŒ ์˜์ƒ์„ ๊ธฐ์ค€์œผ๋กœ ์ž‘์„ฑํ•˜์˜€๋‹ค.

๐Ÿ‘‰ YouTube - Definition of the Possion Process


๋จผ์ € $N(t)$ ๋˜๋Š” $N_t$๋ฅผ ์ •์˜ํ•˜์ž. ์ด๊ฒƒ์€ $t$์‹œ๊ฐ„๊นŒ์ง€ ๋„์ฐฉํ•œ ์‚ฌ๊ฑด์˜ ๊ฐฏ์ˆ˜๋ฅผ ์˜๋ฏธํ•˜๋Š” RV์ด๋‹ค. BP์—์„œ์˜ ์„ฑ์งˆ๋“ค์„ ๋ฐ”ํƒ•์œผ๋กœ <Poisson Process>๋ฅผ ์ž˜ ์ •์˜ํ•ด๋ณด์ž.

1. ๊ฐ time slot์€ ์„œ๋กœ ๋…๋ฆฝ์ด๋‹ค.

Poisson Process๋„ ์ด ์„ฑ์งˆ์„ ๊ฐ€์ง€๋ฏ€๋กœ, ์•„๋ž˜์˜ ๋ช…์ œ๊ฐ€ ์„ฑ๋ฆฝํ•œ๋‹ค.

โ€# of arrivals in disjoint time inteverals are independent.โ€

์ด๊ฒƒ์„ ์ˆ˜์‹์œผ๋กœ ํ‘œํ˜„ํ•˜๋ฉด ์•„๋ž˜์™€ ๊ฐ™๋‹ค.

\[\left( N(t_2) - N(t_1) \right) \perp \left( N(t_4) - N(t_3) \right)\]

2. (Time homogeneity) ๊ฐ time slot์—์„œ arrival์ด ๋ฐœ์ƒํ•  ํ™•๋ฅ ์ด ๋™์ผํ•˜๋‹ค.

๋งˆ์ฐฌ๊ฐ€์ง€๋กœ BP์—์„œ ๊ฐ time slot๋งˆ๋‹ค ๋ชจ๋‘ ํ™•๋ฅ  $p$๋ฅผ ๊ฐ€์กŒ๊ธฐ ๋•Œ๋ฌธ์— Poission Process๋„ ์ด ์„ฑ์งˆ์„ ๊ฐ€์ง„๋‹ค. ์ด๊ฒƒ์„ ๊ธฐ์ˆ ํ•˜๋ฉด,

โ€œ$P(k, \tau)$, the prob. of $k$ arrivals in interval of duration $\tau$ is constantโ€

๊ทธ๋ฆฌ๊ณ  $P(k, \tau)$์— ๋Œ€ํ•ด ์ด๊ฒƒ์„ $k$์— ๋Œ€ํ•ด ๋ชจ๋‘ ๋”ํ•˜๋ฉด, ๊ทธ ํ™•๋ฅ ์˜ ๅˆ์€ 1์ด ๋œ๋‹ค.

\[\sum^{\infty}_{k=0} P(k, \tau) = 1\]

์ˆ˜์—…์—์„  ์ด๊ฑธ ์กฐ๊ธˆ ๋‹ค๋ฅด๊ฒŒ ๊ธฐ์ˆ ํ•œ ๊ฒƒ ๊ฐ™๋‹ค. โ€œThe distribution of $N(t) - N(s)$ only depends on $(t-s)$โ€

\[N(t) - N(s) = N(t-s)\]

3. small interval probability

โ€œ๋‘ arrival์ด ๋™์ผํ•œ ์‹œ๊ฐ„์— ๋™์‹œ์— ๋ฐœ์ƒํ–ˆ๋‹ค.โ€ ์ด๋Ÿฐ ๊ฒฝ์šฐ๋ฅผ ์ƒ๊ฐํ•  ์ˆ˜ ์žˆ์„๊นŒ? ํ˜„์‹ค์—์„œ๋„ ์ด๋Ÿฐ โ€œSame Time, Same place, Same Eventโ€๊ฐ€ ์ผ์–ด๋‚˜๋Š” ๊ฑด ๋ถˆ๊ฐ€๋Šฅํ•˜๋‹ค. Poission Process๋Š” ์ด๋Ÿฐ ๋™์‹œ์— ๋ฐœ์ƒํ•˜๋Š” ์‚ฌ๊ฑด์„ ์—†์• ๊ธฐ ์œ„ํ•ด ์•„์ฃผ ์ž‘์€ interval $\delta$์— ๋Œ€ํ•ด ์•„๋ž˜์™€ ๊ฐ™์ด ์ •์˜ํ•œ๋‹ค.

\[P(k, \delta) \approx \begin{cases} 1 - \lambda \delta & \text{if} \quad k=0 \\ \lambda \delta & \text{if} \quad k=1 \\ 0 & \text{if} \quad k > 1 \end{cases}\]

์ •๋ฆฌํ•˜๋ฉด, ์œ„์™€ ๊ฐ™์€ 3๊ฐ€์ง€ ์กฐ๊ฑด์„ ๋งŒ์กฑํ•œ๋‹ค๋ฉด ์šฐ๋ฆฌ๋Š” ๊ทธ ๊ณผ์ •์„ <Poisson Process>๋ผ๊ณ  ํ•œ๋‹ค!


์ž ๊น ๋‹ค์‹œ <Bernoulli Process>์˜ ์‹œ๊ฐ์œผ๋กœ ๋Œ์•„์™€๋ณด์ž. $[0, t]$ ๊ฐ„๊ฒฉ์„ ๊ฐ€์ง€๋Š” ํ™•๋ฅ  ๋ณ€์ˆ˜ $X$๊ฐ€ ์žˆ๋‹ค๊ณ  ํ•˜์ž. ๊ทธ๋Ÿฌ๋ฉด, ์ด๊ฒƒ์˜ ํ™•๋ฅ ์€

\[\begin{cases} P(X = 1) = \lambda t + o(h) \\ P(X = 0) = 1 - \lambda t + o(h) \end{cases}\]

์ด๋•Œ $X_i$๋ฅผ โ€œ# of buses that arrive in $[t_i, t_{i+1}]$โ€๋ผ๊ณ  ์ •์˜ํ•œ๋‹ค๋ฉด, $X_i$์— ๋Œ€ํ•œ ๋ถ„ํฌ๋Š” Bernoulli Distribution์„ ๋”ฐ๋ฅธ๋‹ค.

\[\begin{cases} P(X = 1) = \lambda \cdot \dfrac{t}{n} + o(h) \\ P(X = 0) = 1 - \lambda \cdot \dfrac{t}{n} + o(h) \end{cases}\] \[X_i \sim \text{Bernoulli}\left( \frac{\lambda t}{n} \right)\]

์ด๋•Œ, $N(t) = X_1 + \cdots + X_n$๋กœ ๋‘”๋‹ค๋ฉด, $N(t)$๋Š” Binomial Distribution $\text{BIN}(n, \lambda t/n)$์„ ๋”ฐ๋ฅด๊ฒŒ ๋œ๋‹ค.

\[X_1 + \cdots + X_n = N(t) \sim \text{BIN}(n, \lambda t/n)\]

์ด๋•Œ, ์šฐ๋ฆฌ๊ฐ€ $n \rightarrow \infty$๋กœ ๋ณด๋‚ด๊ณ  $[t_i, t_{i+1}] \rightarrow 0$๊ฐ€ ๋œ๋‹ค๋ฉด, ์•ž์—์„œ ์–ธ๊ธ‰ํ•œ <Law of Rare event>์— ์˜ํ•ด Binomial Distribution์ด Poisson Distribution์ด ๋œ๋‹ค.

\[\text{BIN}(n, \lambda t/n) \approx \text{POI}(\lambda t)\]

์ •๋ฆฌํ•˜๋ฉด, $N(t)$๋ฅผ ๋ชจ์€ sequence $\{ N(t) : t \ge 0\}$๋Š” <Possion Process>๋‹ค. ๊ทธ๋ฆฌ๊ณ  ๊ฐœ๋ณ„ $N(t)$๋Š” <Poission Distribution>์„ ๋”ฐ๋ฅธ๋‹ค. ๐Ÿคฉ

\[N(t) \sim \text{POI}(\lambda t)\]

Example.

Let $T$ be the time that the 1st bus arrives. What is the distribution of $T$? (We know that the average arrival time is $\lambda$)

์ฃผ์˜ํ•  ์ ์€ ์•ž์—์„œ ์‚ดํŽด๋ณธ <Geometric Distribution>์ฒ˜๋Ÿผ 1st event case๋ฅผ ๊ตฌํ•˜๋Š” ๋ฌธ์ œ์ด์ง€๋งŒ, Sample Space๊ฐ€ ์ด์‚ฐ์ด ์•„๋‹ˆ๋ผ ์—ฐ์†์ธ time axis๋ผ๋Š” ์ ์ด๋‹ค!!

๋จผ์ € cdf $P(T \le t)$๋ฅผ ๊ตฌํ•ด๋ณด์ž. $P(T \le t)$๋ฅผ ์ง์ ‘ ๊ตฌํ•˜์ง€ ๋ง๊ณ , ๋ฐ˜๋Œ€ ์ผ€์ด์Šค์ธ $P(T > t)$๋ฅผ ์ด์šฉํ•ด ์œ ๋„ํ•ด๋ณด์ž.

$P(T > t)$, ์ฆ‰ ๊ธฐ๋‹ค๋ฆฌ๋Š” ์‹œ๊ฐ„ $T$๊ฐ€ $t$๋ณด๋‹ค ์ปค์งˆ ํ™•๋ฅ ์€ ๊ณง $t$ ์‹œ๊ฐ„๊นŒ์ง€ ๋„์ฐฉํ•œ ๋ฒ„์Šค์˜ ์ˆ˜๊ฐ€ 0์ด ๋  ํ™•๋ฅ ๊ณผ ๊ฐ™๋‹ค. ์ฆ‰, $N(t) = 0$์˜ ํ™•๋ฅ ๊ณผ ๊ฐ™๋‹ค. ๋”ฐ๋ผ์„œ,

\[P(T > t) = P(N(t) = 0) = e^{-\lambda t} \frac{(\lambda t)^0}{0!} = e^{-\lambda t}\]

๋”ฐ๋ผ์„œ, $P(T \le t) = 1 - e^{-\lambda}$์ด๋‹ค. ์ด๊ฒƒ์„ ๋ฏธ๋ถ„ํ•˜๋ฉด pdf $f(x)$๋ฅผ ์–ป์„ ์ˆ˜ ์žˆ๋‹ค.

\[\frac{d}{dt} P(T \le t) = \frac{d}{dt} (1 - e^{-\lambda t}) = \lambda \cdot e^{-\lambda t}\]

๋’ค์—์„œ ๋‹ค๋ฃจ๊ฒ ์ง€๋งŒ, ์œ„์™€ ๊ฐ™์€ pdf๋ฅผ ๊ฐ€์ง€๋Š” continuous distribution์„ <Exponential Distribution>์ด๋ผ๊ณ  ํ•œ๋‹ค.


์ด๋ฒˆ ํฌ์ŠคํŠธ์—์„œ ๋‹ค๋ฃฌ <Poisson Distribution>์„ ๋์œผ๋กœ ๊ต์žฌ์—์„œ ๋‹ค๋ฃจ๋Š” ๋ชจ๋“  ์ด์‚ฐ ํ™•๋ฅ  ๋ถ„ํฌ๋ฅผ ์‚ดํŽด๋ณด์•˜๋‹ค. ๋‹ค์Œ ํฌ์ŠคํŠธ๋ถ€ํ„ฐ๋Š” ์—ฐ์† RV๊ฐ€ ๊ฐ–๋Š” <์—ฐ์† ํ™•๋ฅ  ๋ถ„ํฌ; Continuous Distribution>์— ์‚ดํŽด๋ณด๊ฒ ๋‹ค.