ПАНДЕМИЯ МЫСАЛЫНДА ЭПИДЕМИОЛОГИЯЛЫҚ ӨРШУЛЕРДІ БОЛЖАУ ЖӘНЕ ДЕНСАУЛЫҚ САҚТАУДЫ БАСҚАРУ ҮШІН МАШИНАЛЫҚ ОҚЫТУҒА НЕГІЗДЕЛГЕН ИНТЕЛЛЕКТУАЛДЫҚ ЖҮЙЕЛЕР

Аңдатпа

This study proposes an intelligent decision-support system for public-health authorities, combining a rule-based risk-alert layer with two machine-learning modules. Model A (XGBoost gradient boosting) solves a regression task, forecasting confirmed COVID-19 cases for every nation two weeks ahead. Model B (gradient boosting) uses temporal features (lags, rolling means) together with a hospital-pressure indicator derived from the case-fatality ratio (CFR) as a proxy in the absence of direct ICU data, classifying each weekly observation as ICU-Critical or Normal. Experiments used the open Johns Hopkins University CSSE dataset (201 nations, January 22, 2020 – March 9, 2023; 32,964 country–week observations). On an independent test set (n = 5,542) Model A obtained R² = 0.611, MAE = 19,792 cases per week, and RMSE = 118,513; Model B obtained AUC = 0.986, accuracy 0.95, and F1 = 0.91 for the ICU-Critical class. The two models form a four-level warning system. The scientific novelty comprises the CFR-proxy methodology for estimating inpatient burden without open ICU data, a single model that generalizes across 201 countries, and the integration of forecasting and classification into one recommendation system applicable to tactical bed-capacity planning and early warning.