A master's thesis at Datalab
We, Joris and Luc, are students on the master's programme Applied Data Science at Utrecht University. Commissioned by Datalab and in collaboration with De Administratie, we accepted the challenge of developing a predictive model for cashflow. In this blog we are happy to share our findings and the lessons we learned during our research.
Predicting cashflow: is that even possible?
With our particular interest in data analysis and machine learning, we were enthusiastic about working on this project. Our goal was to gain valuable insights and create a model that could support accountants and financial controllers in predicting cashflow. We used advanced machine learning algorithms, including XGBoost, Random Forest, SVM and MLP, to explore different approaches and identify the best predictive models.
Throughout the research, we collaborated intensively with Datalab to explore the datasets and optimise the models. We carried out various experiments, analysing and comparing the performance of the models. Enriching/optimising and achieving uniformity in datasets for such predictions is the primary outcome we drew.
No accurate prediction, but a valuable lesson
These findings are particularly valuable. They underscore the importance of diligence when using machine learning models for financial predictions. It is essential to recognise that the accuracy of predictions depends on the availability and quality of the underlying data. In our case, there were various complicating factors in the dataset that made predicting cashflow difficult. Although our findings do not enable direct application for accountants and financial controllers, this research offers valuable insights for further improving data quality and exploring additional factors that may influence cashflow predictions. It also illustrates the importance of continuous research and development in the field of financial models and machine learning in the accounting sector.
We would like to thank Datalab for the opportunity to work on this research and gain valuable experience. Although we could not accurately predict cashflow with the current dataset, we are convinced that De Administratie will predict cashflow in the future with the help of machine learning.
We hope this blog has provided insight into our research process and the challenges we encountered.
Thank you for reading and goodbye! Joris & Luc