The Influence of Twitter Sentiment Analysis on Predicting Tuberculosis Cases in Indonesia Using CNN-GRU with Copernicus Air Pollution Data

Tuberculosis (TB) is an infectious disease and a major global health threat, with Indonesia ranking as the country with the second-highest number of TB cases. The COVID-19 pandemic led to a decline in TB case reporting. Social stigma toward TB patients has also emerged in Indonesian communities, causing individuals to hesitate to disclose their illness to others and contributing to the daily rise in cases. These responses are often expressed on social media platforms such as Twitter (X). This study aims to develop a prediction model that incorporates public sentiment related to TB on Twitter and examines how this sentiment affects the accuracy of forecasting future trends in TB cases. The results show that the baseline Convolutional Neural Network-Gated Recurrent Unit (CNN-GRU) model demonstrates strong performance and stability with an 80:10:10 data split and a sequence length of 60, achieving an R² of 0.97, a Mean Absolute Percentage Error (MAPE) of 0.80 percent, a Root Mean Square Error (RMSE) of 20.518, and a Mean Absolute Error (MAE) of 18.22 on the test data. Adding the Twitter sentiment feature subjectivity_rollingwindow365 does not consistently improve performance, especially when the TB and air pollution data are already robust. This addition only enhances accuracy in baseline models that are less precise. These findings indicate that sentiment can serve as a supplementary signal when the primary data are insufficient, although it provides limited benefit for prediction models that already perform well.