Modelo para predecir las deudas de los contribuyentes con alta probabilidad de no pago en la administración tributaria

Abstract

One of the main processes in tax administrations is debt management. The objective of this process, among others, is to recover the economic resources declared on the statement made by taxpayers. Due to limitations in tax administrations, such as, staff, tools, time, etc., tax administrations seek to recover debts in the early stages of control, where the cost of collection is lower than in the subsequent stages. In order to optimize the debt management process and contribute to decision-making, this Master's Thesis proposes to carry out a data mining process to develop a model to predict debts of taxpayers with high probability of non-payment. The extraction of knowledge as a result of the data mining process identifies high-risk debts for the debt management process. In addition, our study also allows to develop a survival analysis of the debts of a dataset from a tax administration using machine learning techniques. Three groups of tax debtors with similar payment behavior were identified and a performance of up to 90% was obtained to estimate the payment time of taxpayers. Concordance index metric was used to measure the performance.