Application of Time Series and Machine Learning Approaches to Prediction of the Amount of Administrative Enforcement Agency in Taiwan

• Main Authors: HO, FENG-LIN, 何鳳苓
• Other Authors: SHAO, YUEHJEN E.
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/ty835x

碩士 === 輔仁大學 === 統計資訊學系應用統計碩士在職專班 === 106 === The sub-bureaus of Taiwan's administrative executive offices have been established in 2001. As long as the creditor’s right to the debts is a public agency, they can be transferred to the administrative enforcement agencies. The amount of money collected is an important performance indicator for administrative execution. People and public opinion have focused on whether the implementation of the executive branch of government has been more effective than the predecessor's court and has grown steadily since the establishment of the administrative executive authority. Whether the illegally concealed or evasive tax delinquent households demonstrate the government’s determination and determination to implement public power, strengthen clean-up implementation, and increase national treasury revenue. Therefore, it is an important issue to discuss the amount of administrative enforcement agencies to collect the amount. This study uses the autoregressive integrated moving average (ARIMA), artificial neural network (ANN), and multivariate adaptive regression splines (MARS) methods to predict four types of amount of the administrative enforcement agency, financial and tax, health insurance, fines, and fees. The mean absolute percentage error (MAPE) was used to compare the accuracy of various forecasting models. The results indicate that the MARS models have the better performance for the predictions of financial and tax, fines cases, and the ARIMA methods have better performance for predictions of health insurance and fees case.