| Popis: |
The COVID-19 pandemic that started in 2020 had a negative influence on many businesses worldwide, leading to financial uncertainties within different business areas. As a result, many companies were unable to pay their monthly electricity usage on time to energy suppliers. This created a continuous loop of accumulating financial liabilities for many companies that rely on services from others, while not being able to pay timely. To remain competitive during such difficult times with focus on limiting financial losses, energy suppliers rely on the power of data to optimize business to business customer processes. One method is by using machine learning to improve current debtor models that predict the probability of default. This paper presents an analysis of customer payment behaviour data using advanced machine learning approaches, such as XGBoost, LightGBM and LSTM, to anticipate the probability of default as a main business delivery. Results of this work show that LSTM outperforms the current debtors models as it utilizes the persisting trends changes in the second half of 2020 and focuses on the most significant customer segmentation groups such as high/low monthly payers and high/low credit risks scores. |
