Abstract
This paper develops a prediction study of a group of small businesses which have a higher risk of non-compliance with taxation obligations. These businesses have been selected for a pre-emptive SMS reminder campaign and prediction models are used to predict the probability of on-Time payment. Through experiments on a real world taxation debt dataset, it is found that the XGBoost algorithm significantly outperforms random forest, decision tree and logistic regression algorithms. The variables showing the largest explanatory power are related to debt amount. Second and subsequent SMS messages make a negligible contribution to the probability of payment. The XGBoost explainer is also used to delve further into the inner workings of the algorithm. Copyright © 2018 IEEE.
| Original language | English |
|---|---|
| Title of host publication | Proceedings of 2018 5th International Conference on Behavioral, Economic, and Socio-Cultural Computing, BESC 2018 |
| Place of Publication | Danvers, MA |
| Publisher | IEEE |
| Pages | 19-23 |
| ISBN (Electronic) | 9781728102078 |
| DOIs | |
| Publication status | Published - Jul 2018 |
Citation
Cao, G., Downes, A., Khan, S., Wong, W., & Xu, G. (2018). Taxpayer behavior prediction in SMS campaigns. In Proceedings of 2018 5th International Conference on Behavioral, Economic, and Socio-Cultural Computing, BESC 2018 (pp. 19-23). IEEE. https://doi.org/10.1109/BESC.2018.8697317Keywords
- Taxpayer behavior
- Debt collection
- SMS
- Prediction
- XGBoost