Artificial Intelligence (AI) Driven Forecasting of Value Added Tax: A Case Study of the Tanzania Revenue Authority

This study set out to assess the applicability of Artificial Intelligence (AI) models in forecasting Value Added Tax revenue, with a focus on the Tanzania Revenue Authority. In assessing the applicability, the historical trends and performance of VAT collection were used to develop AI models to forecast monthly VAT revenue, thereafter providing recommendations for improving VAT forecasting through the adoption of AI technologies. The study utilised time series for the period of January 2021 to December 2024, constructing a dataset covering 48 months. The variables under the data set are: VAT revenue statistics, which provide detailed transactional records of VAT collections across three main categories: imports, purchases, and sales. The granular daily entries are aggregated to obtain monthly totals, creating a coherent time series that is suitable for statistical modelling and forecasting purposes. Other variables include the Consumer Price Index (CPI), GDP and inflation rate. Five forecasting models, SARIMAX, Elastic-Net, XGBoost, CNN-LSTM, and SNaive, were basically applied in predicting VAT revenue. All models under testing performed at different levels, with XGBoost performing better with the lowest MAE TZS 150 billion, RMSE TZS 205 billion and MAPE 275%, demonstrating its effectiveness in capturing non-linear patterns and handling outliers in VAT revenue data. The Elastic-Net had an MAE of TZS 159 billion, a MAPE of 933% and a moderate RMSE of TZS 2.49 billion, pinpointing its limitations in relative accuracy and large error management. SARIMAX and SNaive slightly underperform with MAE of TZS 8.42 billion and TZS 904, respectively, and MAPE of 344% and 322%. The CNN-LSTM underperforms with an MAE of TZS 787 billion, an RMSE of TZS 2.54 and a MAPE of 873%, suggesting challenges in overfitting or pattern complexity. By combining methodological rigour, domain-specific feature engineering, and robust validation, these models offer a credible pathway toward more accurate, timely, and actionable fiscal forecasts. Therefore, the study results revealed that AI-based methods, particularly ensemble modelling, can be substantially applied in VAT revenue forecasting for TRA.

Authors

Institutions

Publication Details

Journal
International Journal of Research Publications
Published
2026-11-13
DOI
https://doi.org/10.47119/ijrp10020811120269900
Primary Topic
Forecasting Techniques and Applications
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Artificial Intelligence (AI) Driven Forecasting of Value Added Tax: A Case Study of the Tanzania Revenue Authority

Miraji H. Kasunsumo, Aweso H. Aweso
2 links
International Journal of Research Publications
Forecasting Techniques and Applications
article

Artificial Intelligence (AI) Driven Forecasting of Value Added Tax: A Case Study of the Tanzania Revenue Authority

Miraji H. Kasunsumo, Aweso H. Aweso
article en

Abstract

This study set out to assess the applicability of Artificial Intelligence (AI) models in forecasting Value Added Tax revenue, with a focus on the Tanzania Revenue Authority. In assessing the applicability, the historical trends and performance of VAT collection were used to develop AI models to forecast monthly VAT revenue, thereafter providing recommendations for improving VAT forecasting through the adoption of AI technologies. The study utilised time series for the period of January 2021 to December 2024, constructing a dataset covering 48 months. The variables under the data set are: VAT revenue statistics, which provide detailed transactional records of VAT collections across three main categories: imports, purchases, and sales. The granular daily entries are aggregated to obtain monthly totals, creating a coherent time series that is suitable for statistical modelling and forecasting purposes. Other variables include the Consumer Price Index (CPI), GDP and inflation rate. Five forecasting models, SARIMAX, Elastic-Net, XGBoost, CNN-LSTM, and SNaive, were basically applied in predicting VAT revenue. All models under testing performed at different levels, with XGBoost performing better with the lowest MAE TZS 150 billion, RMSE TZS 205 billion and MAPE 275%, demonstrating its effectiveness in capturing non-linear patterns and handling outliers in VAT revenue data. The Elastic-Net had an MAE of TZS 159 billion, a MAPE of 933% and a moderate RMSE of TZS 2.49 billion, pinpointing its limitations in relative accuracy and large error management. SARIMAX and SNaive slightly underperform with MAE of TZS 8.42 billion and TZS 904, respectively, and MAPE of 344% and 322%. The CNN-LSTM underperforms with an MAE of TZS 787 billion, an RMSE of TZS 2.54 and a MAPE of 873%, suggesting challenges in overfitting or pattern complexity. By combining methodological rigour, domain-specific feature engineering, and robust validation, these models offer a credible pathway toward more accurate, timely, and actionable fiscal forecasts. Therefore, the study results revealed that AI-based methods, particularly ensemble modelling, can be substantially applied in VAT revenue forecasting for TRA.

International Journal of Research PublicationsVol. 208(1)
Government of Tanzania (TZ)
Forecasting Techniques and Applications
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon