2nd/10_Wiki/Topics/AI_and_ML/Epidemiological-Modeling.md

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

id	title	category	status	canonical_id	aliases	duplicate_of	source_trust_level	confidence_score	verification_status	tags	raw_sources	last_reinforced	github_commit	tech_stack
wiki-2026-0508-epidemiological-modeling	Epidemiological Modeling	10_Wiki/Topics	verified	self	SIR model SEIR compartmental model disease modeling R0 agent-based epi	none	A	0.95	applied	epidemiology modeling sir public-health simulation forecasting covid		2026-05-10	pending	language framework Python scipy / NetworkX / Mesa / NumPyro

Epidemiological Modeling

매 한 줄

"매 disease 의 population 의 spread 의 model". 매 SIR / SEIR (compartmental), 매 ABM (network), 매 statistical (R_t estimate). 매 R0, herd immunity, 매 NPI effect. 매 modern: 매 ML forecasting + 매 mobility data + 매 Bayesian inference.

매 핵심

매 compartmental

• SIR: Susceptible → Infected → Recovered.
• SEIR: + Exposed.
• SIRS: 매 immunity 의 wane.
• SIRD: + Dead.
• MSEIR: + Maternal immunity.

매 key parameter

• R0: 매 basic reproduction number.
• R_t: 매 effective (time-varying).
• β: 매 transmission rate.
• γ: 매 recovery rate.
• R0 = β / γ.
• Herd immunity: 매 1 - 1/R0.

매 method

• ODE: 매 mean-field, deterministic.
• Stochastic (Gillespie): 매 small population.
• ABM: 매 individual + network.
• Statistical: 매 R_t from cases.
• ML / DL: 매 forecasting.

매 응용

1. Pandemic forecast: 매 COVID, flu.
2. Vaccination strategy: 매 priority.
3. NPI effect: 매 lockdown, mask.
4. Travel ban: 매 border.
5. Hospital capacity: 매 ICU.
6. Animal: 매 livestock disease.

💻 패턴

SIR (ODE)

import numpy as np
from scipy.integrate import odeint

def sir(state, t, beta, gamma, N):
    S, I, R = state
    dS = -beta * S * I / N
    dI = beta * S * I / N - gamma * I
    dR = gamma * I
    return [dS, dI, dR]

t = np.linspace(0, 200, 1000)
N = 1_000_000
sol = odeint(sir, [N - 1, 1, 0], t, args=(0.4, 0.1, N))

SEIR

def seir(state, t, beta, sigma, gamma, N):
    S, E, I, R = state
    dS = -beta * S * I / N
    dE = beta * S * I / N - sigma * E
    dI = sigma * E - gamma * I
    dR = gamma * I
    return [dS, dE, dI, dR]

Stochastic Gillespie

def gillespie_sir(N, I0, beta, gamma, t_max=200):
    S, I, R = N - I0, I0, 0
    t = 0
    history = [(0, S, I, R)]
    while I > 0 and t < t_max:
        a1 = beta * S * I / N  # 매 infection rate
        a2 = gamma * I         # 매 recovery rate
        a0 = a1 + a2
        if a0 == 0: break
        tau = np.random.exponential(1 / a0)
        t += tau
        if np.random.rand() < a1 / a0:
            S -= 1; I += 1
        else:
            I -= 1; R += 1
        history.append((t, S, I, R))
    return history

R_t estimation (EpiEstim-style)

def estimate_rt(case_counts, gen_time=5, window=7):
    """매 simple Cori method approximation."""
    rt = []
    for t in range(window, len(case_counts)):
        recent = case_counts[t - window:t]
        infectees = sum(recent[-(window - i)] * np.exp(-(window - i) / gen_time) for i in range(window))
        rt.append(case_counts[t] / max(infectees, 1))
    return rt

Network ABM (NetworkX)

import networkx as nx
import random

def network_sir(G, beta=0.1, gamma=0.05, init_infected=5, steps=100):
    state = {n: 'S' for n in G.nodes()}
    for n in random.sample(list(G.nodes()), init_infected): state[n] = 'I'
    
    history = []
    for _ in range(steps):
        new_state = state.copy()
        for n in G.nodes():
            if state[n] == 'I':
                if random.random() < gamma: new_state[n] = 'R'
                for nb in G.neighbors(n):
                    if state[nb] == 'S' and random.random() < beta:
                        new_state[nb] = 'I'
        state = new_state
        history.append({'S': sum(1 for v in state.values() if v == 'S'),
                        'I': sum(1 for v in state.values() if v == 'I'),
                        'R': sum(1 for v in state.values() if v == 'R')})
    return history

G = nx.barabasi_albert_graph(1000, 3)
hist = network_sir(G)

Vaccination intervention

def vaccinate(G, strategy='degree', frac=0.3):
    n = int(G.number_of_nodes() * frac)
    if strategy == 'degree':
        targets = sorted(G.nodes(), key=G.degree, reverse=True)[:n]
    elif strategy == 'random':
        targets = random.sample(list(G.nodes()), n)
    elif strategy == 'betweenness':
        bc = nx.betweenness_centrality(G)
        targets = sorted(bc, key=bc.get, reverse=True)[:n]
    return set(targets)

Bayesian SIR (NumPyro)

import jax.numpy as jnp
import numpyro
import numpyro.distributions as dist

def bayesian_sir(observed_cases, N):
    beta = numpyro.sample('beta', dist.Uniform(0.1, 1.0))
    gamma = numpyro.sample('gamma', dist.Uniform(0.05, 0.3))
    sigma_obs = numpyro.sample('sigma', dist.HalfNormal(10))
    
    # 매 simulate SIR (deterministic given params)
    predicted = simulate_sir_jax(beta, gamma, N, len(observed_cases))
    numpyro.sample('obs', dist.Normal(predicted, sigma_obs), obs=observed_cases)

Mobility-aware (commute)

def metapopulation_sir(populations, mobility, beta, gamma):
    """매 multi-region 의 commute matrix."""
    n = len(populations)
    S = populations.copy()
    I = np.zeros(n); I[0] = 10
    R = np.zeros(n)
    for _ in range(200):
        # 매 effective infectious in each region with commuters
        I_eff = mobility @ I
        new_I = beta * S * I_eff / populations
        new_R = gamma * I
        S -= new_I; I += new_I - new_R; R += new_R
    return S, I, R

NPI effect (intervention)

def sir_with_npi(t, beta, gamma, lockdown_start, lockdown_end, lockdown_factor=0.3):
    if lockdown_start <= t <= lockdown_end:
        beta = beta * lockdown_factor
    return beta, gamma

Forecast (ML over compartmental)

import xgboost as xgb

def forecast_cases(history, window=14, horizon=7):
    """매 ML residual on top of SIR."""
    sir_pred = sir_simulate(history)
    residuals = history - sir_pred
    
    X = sliding_window(residuals, window)
    y = residuals[window:window + len(X)]
    model = xgb.XGBRegressor().fit(X, y)
    
    return sir_pred[-horizon:] + model.predict(X[-1:])[:horizon]

매 결정 기준

상황	Approach
Quick estimate	SIR ODE
Latency disease	SEIR
Small outbreak	Gillespie stochastic
Network spread	ABM on graph
Real-time R_t	Cori / EpiEstim
Forecast	Compartmental + ML residual
Spatial	Metapopulation + mobility

기본값: 매 SEIR baseline + 매 Bayesian inference + 매 mobility data + 매 ML residual + 매 NPI scenario analysis.

🔗 Graph

• 변형: SEIR · Compartmental-Model
• Adjacent: Bayesian Inference

🤖 LLM 활용

언제: 매 outbreak. 매 vaccination plan. 매 hospital capacity. 언제 X: 매 individual diagnosis (different domain).

❌ 안티패턴

• R0 의 over-trust: 매 heterogeneity 의 ignore.
• Mean-field at small scale: 매 stochastic 의 use.
• No data calibration: 매 toy.
• Forecast far horizon: 매 uncertainty 의 hide.
• Single model: 매 ensemble 의 prefer.

🧪 검증 / 중복

• Verified (Anderson & May, Cori 2013, COVID-19 modeling literature).
• 신뢰도 A.

🕓 Changelog

날짜	변경
2026-04-20	Auto-reinforced
2026-05-08	Phase 1
2026-05-10	Manual cleanup — SIR / SEIR / Gillespie / ABM / Bayes / NPI code