Overdue invoice forecasting and data mining

Abstract

The account receivable is one of the main challenges in the business operation. With poor management of invoice to cash collection process, the over due invoice may pile up, and the increasing amount of unpaid invoice may lead to cash flow problems. In this thesis, I addressed the proactive approach to improving account receivable management using predictive modeling. To complete the task, I built supervised learning models to identity the delayed invoices in advance and made recommendations on improving performance of order to cash collection process. The main procedures of the research work are data cleaning and processing, statistical analysis, building machine learning models and evaluating model performance. The analytical and modeling of the study are based on the real-world invoice data from a Fortune 500 company. The thesis also discussed approaches of dealing with imbalanced data, which includes sampling techniques, performance measurements and ensemble algorithms. The invoice data used in this thesis is imbalanced, because on-time invoice and delayed invoice classes are not approximately equally represented. The cost sensitivity learning techniques demonstrates favorable improvement on classification results. The results of the thesis reveal that the supervised machine learning models can predict the potential late payment of invoice with high accuracy.