Analysis of Tuberculosis Contact Investigation Using the Integrated Association Rules Method of Time Series Forecasting

Abstract

This study integrates association rule mining and time series forecasting to analyze tuberculosis contact investigation data more accurately. Inter-variable relationships were explored using the FP-Growth algorithm, resulting in seven association rules. Subsequently, three forecasting algorithms (Random Forest, XGBoost, and LightGBM) were compared to determine the most accurate predictive model. Results showed that Random Forest outperformed the others with the lowest error values. The rule {Cough, Household Contact} - {Referred} yielded the highest prediction value of 273 with a MAE of 0.001. Findings indicate that individuals aged 25-65, particularly the elderly, who experience cough, shortness of breath, and are smokers living with TB patients, are at high risk of becoming new cases. This approach contributes to tuberculosis control programs by systematically identifying patterns and projecting potential cases.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint