Mathematical modelling of time series data for tuberculosis notified cases in India using neural networks models: CNN, NNAR, and ANFIS models integrated with wavelets

Abstract

Tuberculosis (TB) remains a persistent and critical public health challenge in India, contributing significantly to the global disease burden. Despite ongoing control measures, seasonal surges and underreporting continue to hinder timely intervention and resource allocation. There is an urgent need for accurate, data-driven forecasting tools to predict TB case trends and enable proactive healthcare planning. This study addresses this necessity by employing advanced Artificial Intelligence-based time series models, specifically NNAR, ANFIS, CNN, and their wavelet-integrated variants, to forecast TB notifications in India using data from the NIKSHAY database (2017–2022). By capturing the seasonal and trend dynamics inherent in TB cases, the study supports data-informed decision-making for public health authorities. The results demonstrate that wavelet-enhanced models significantly enhance predictive accuracy. Notably, the NNAR-Db8L2 model reduces forecasting errors by over 39%, while the CNN-D8L5 and ANFIS-Db8L6 models also show marked improvements, proving effective in modelling complex seasonal patterns. These findings emphasize the demand for hybrid AI models in disease surveillance and their potential to inform timely, evidence-based TB control strategies.