azizadeboye/rf-barima-tb-forecasting-south-africa

Forecasting tuberculosis (TB) incidence among children under 5 in South Africa using a hybrid Random Forest-Bayesian ARIMA model # RF-BARIMA TB Forecasting in South Africa This repository accompanies the manuscript: **Title:** *Machine Learning Forecasting Model of Tuberculosis Cases Among Children in South Africa* ## Abstract Globally, children under 15 years old represent ~11% of all TB cases. The burden in those under 5 years in South Africa remains underexplored. This study presents a novel hybrid Random Forest-Bayesian ARIMA (RF-BARIMA) model to forecast TB incidence in Eastern Cape Province from 2010 to 2019. The RF-BARIMA model demonstrated superior forecasting performance over traditional models. Forecasts estimate an average of 0.4122 TB cases/month in 2022, suggesting a potential reduction of 1670.85 mean cases. These results highlight potential under-reporting and call for urgent improvements in surveillance and public health interventions. Forecasting tuberculosis (TB) incidence among children under 5 in South Africa using a hybrid Random Forest-Bayesian ARIMA model --- ## Forecasting Pipeline The modelling workflow includes: - Data ingestion and preprocessing - Model training using ARIMA, Bayesian ARIMA (via `bayesmodels`), Prophet, TBATS, and MARS - Hybridisation with XGBoost and model calibration using `modeltime` - Forecast generation and accuracy evaluation - Final forecasting for 3 years beyond the dataset ## Packages Used library(tidyverse) library(lubridate) library(modeltime) library(bayesmodels) library(forecastHybrid) library(timetk) library(catboost) library(prophet) library(tbats) library(recipes) library(parsnip) library(rsample)