Prediction of Late Payments Using Decision Tree-Based Algorithms

Abstract

The Invoice-to-Cash process is essential for the financial stability of any company, in view of the collection of accounts receivable as its main activity. However, despite its importance, the billing stage is usually processed manually, which causes contact with all customers at fixed intervals, even if some have always paid on time. Thus, the work explores data mining techniques with machine learning, aiming to optimize the collection process through prediction of invoice payments. For this, eight algorithms based on decision tree were used, applied in three steps: (i) identify invoices with payment on time or late; (ii) identify among the overdue invoices, payment in the due date month or later; and (iii) predict among the overdue invoices, how many days of delay they will have beyond the due month. Finally, through the results obtained from the best models for each step, was obtained an average precision of 81.85%, 85.63% and 73.98%, respectively.