A hybrid model to mitigate data gaps and fluctuations in tax revenue forecasting

Abstract

This study addresses the critical challenge of advancing tax revenue forecasting models to effectively handle distinctive data gaps and inherent fluctuations in tax revenue data. These challenges are evident in Lampung Province, Indonesia, where limited temporal granularity and non-linear variability hinder accurate fiscal planning. Despite advancements in statistical, machine learning, and hybrid approaches, existing models often fall short in simultaneously managing these challenges. A hybrid model integrating random forest regressors for data interpolation and Long Short-Term Memory for capturing complex temporal patterns was proposed. The model was evaluated, achieving an R² of 0.86, root mean squared error (RMSE) of 9.65 billion, and mean absolute percentage error (MAPE) of 3.49%. Although the model has limitations in generalizing to unseen data, the results demonstrate that it outperforms existing forecasting models regarding accuracy and reliability. Integrating random forest regressors and long short-term memory delivers a tailored solution to the complexities of tax revenue forecasting, contributing to fiscal forecasting and setting a foundation for further exploration into hybrid approaches.