Nowadays, making financial decisions and evaluating loan applications is a complex and sensitive process. Cash flow data, which shows the financial risk status of businesses, plays a key role in evaluating loan applications. Cash flow data, which shows the financial risk status of businesses, plays a key role in evaluating loan applications. Guiding business managers in making strategic decisions and managing financial risks, quarterly data provides a detailed timeline of business performance and helps identify seasonal changes. A detailed analysis using machine learning algorithms evaluates the performance of different models built to compare businesses quarters in the loan classification process and highlights the role of cash flow data in the process. It was aimed to create effective algorithms by taking into account the suitability of the quarterly data between 2018 and 2022 of the 282 companies used in the study, and to provide a unique approach in the field of evaluating these algorithms with information criteria. The model performances of the quarters are very close to each other and a high success rate is obtained. Therefore, it was observed that quarterly periods did not make a significant difference in model performance. The model created for the 2nd quarter of 2019 was selected as the best model with 99% accuracy and 99% F1 value. It was also determined that the selection of variables with high accuracy rates in the models established for each quarter is important in terms of predicting financial risk.
credit classification machine learning algorithms cash flow statements risk ınformation criteria
Primary Language | English |
---|---|
Subjects | Machine Learning (Other), Econometric and Statistical Methods, Economic Models and Forecasting, Time-Series Analysis, Statistical Data Science, Theory of Sampling, Risk Analysis, Applied Statistics |
Journal Section | Articles |
Authors | |
Early Pub Date | March 27, 2024 |
Publication Date | March 27, 2024 |
Submission Date | December 12, 2023 |
Acceptance Date | February 26, 2024 |
Published in Issue | Year 2024 |
INDEXING