FORECASTING TUBERCULOSIS INFECTIONS USING ARIMA AND HYBRID NEURAL NETWORK MODELS AMONG CHILDREN BELOW 15 YEARS IN HOMA BAY AND TURKANA COUNTIES, KENYA

Abstract

Tuberculosis (TB) among children under the age of 15 is a significant public health problem, particularly in resource-constrained settings and is among top ten most dangerous causes of death worldwide, and ranks among the top five most lethal infectious agents in Kenya. However, the real burden of tuberculosis among children in Kenya is unclear. In modelling infectious diseases, Autoregressive Integrated Moving Average (ARIMA) and hybrid ARIMA models have been widely used. However, few studies in Kenya have utilized ARIMA or hybrid ARIMA models to model infectious diseases. This study sought to forecast TB infections in children under the age of 15 Homa Bay and Turkana Counties in Kenya using ARIMA and hybrid neural network models and specifically sought to compare the; performance of the models in predicting TB notification cases, accuracy produced by the models, and the forecasted temporal trends of TB notification cases among children below 15 years. The study hypothesized that the hybrid ARIMA-ANN model yields more accurate predictions and forecasts. The study used monthly TB confirmed cases reported for Homa Bay and Turkana Counties between 2012 and 2021. The ARIMA model was chosen using the Akaike Information and Bayesian Information Criteria. The ANN model was developed using the Multi-Layer Perceptrons (MLPs) three-layer feed-forward architecture. The hybrid ARIMA model was developed by combining the fitted cases using the ARIMA model and the residuals from the ANN. The hybrid ARIMA model (ARIMA-ANN) outperformed the single ARIMA(0,0,1,1,0,1,12) and ANN (1,1,2)[12] models in terms of predictive and forecast accuracy. The hybrid ARIMA model outperformed the ANN (1,1,2)[12] and ARIMA (0,0,1,1,0,1,12) models in terms of prediction accuracy, p<0.001. In Homa Bay and Turkana Counties, the 12-month predicted TB incidence of 175 to 198 infections per 100,000 children in 2022. The hybrid ARIMA model provides superior prediction accuracy and forecast performance. The findings of this study suggest that TB cases in children are underreported, and that the incidence of TB in children may be greater than previously assumed. Tuberculosis monitoring data needs to be re-evaluated in order to comprehend current inadequacies. To get the TB battle back on track, it is critical to reallocate critical resources to the National TB program.