Cash Collection Forecast using ML for B2B company

Title: Cash Collection Forecast using ML for B2B company Author: Ing. Jan Kříž Department: Informatics and Statistics Supervisor: Ing. Pavel Zimmermann PhD. This thesis aims to use machine learning, specifically regression algorithms, to predict the payment dates of account receivables transactions, including the payment date for invoices and the application date for credit memo transactions. Companies can benefit from this as it predicts the total amount of cash inflows they can receive every week and impacts cash flow and cash management decisions. The extent to how accurate the prediction (when a customer is going to pay for an invoice or when a customer is going to apply the credit memo) is, depends mainly on the features used for the model training and the machine learning algorithm. This thesis describes the entire process, from data loading, cleaning, and feature engineering, to the application of regression and time series models for predictions. The exploratory analysis and data preparation were performed with the use of SQL, while the advanced part of machine learning was built in Python programming language. As the outcome of this project is relevant to business users, models were evaluated in an easy-to-understand process, bearing the end-user in mind for those without statistical and technical backgrounds. The ultimate goal of this project is to improve the cash flow forecast process for the Treasury Department and develop a decision support tool, which will cover all their needs.