Prognosticate Bankruptcy: Machine Learning to Convey the Chance of Distress

  • Abdul Kadar Muhammad Masum, Mohammed Abdur Rahman, Erfanul Hoque Bahadur, Mohammed Shamsul Alam, Md. Kalim Ullah Chowdhury

Abstract

Contemporary widespread financial crisis and escalate of systemic risk in elevated economic expenses have attracted attention to companies' solvency monitoring processes. Predicting bank failures is therefore essential to prevent and/or reduce deleterious effects on the economic system. Inevitably, this study seeks to use prodigious machine learning approaches on bank failure prediction in the United States cases. In the timespan between 1993 and 2018, a total of 24 years, we collected Federal Deposit Insurance Corporation call reports. In order to observe the CAMELS ratios, 23 variables were obtained for each bank from its financial statements and 20 financial ratios converted into subsequent variables. The dataset holds 924,953 occurrences in total. After preparing the bankruptcy dataset, we preprocessed amassed data. Optimizing hyper-parameters with Grid search paradigm, we found out that Artificial Neural Network (ANN) with 98.35% accuracy on the validation dataset is the best classifier performing in the classification among other data mining models. After that, training on the instances in the time horizon of 1993 to 2010 with the pre-trained ANN, we predicted on the instances of each year from 2011 to 2018. Then, learning the ANN with the optimized hyper-parameters, we predicted the performance of the ANN for every predictions dataset separately. The forecasting result of eight years prior to bankruptcy happened was predominant other than any previous work.  The exactness of this entire execution system was 99.13%, 99.06%, 98.78%, 98.67%, 98.72%, 98.59%, 98.44%, and 98.21% respectively. This strategy would raise red flags for financial institutions, which are expected to be collapsing prematurely. Expedient measures can be taken to closely track so that insolvency and contagion cannot be stopped or at least to reduce their costs for the regulatory authority and the economy. These findings also report to highlight that unsuccessful companies focus more on immobilized lending and have additional clauses. We exhibited a more prominent method with outperformance in the event of bankruptcy prediction and rationality of financial distress. In the event that any US bank may prepare this suppositional scheme for predicting probability of collapsing prior to eight years back.

Published
2020-11-05
How to Cite
Abdul Kadar Muhammad Masum, Mohammed Abdur Rahman, Erfanul Hoque Bahadur, Mohammed Shamsul Alam, Md. Kalim Ullah Chowdhury. (2020). Prognosticate Bankruptcy: Machine Learning to Convey the Chance of Distress . International Journal of Advanced Science and Technology, 29(04), 10777–10795. Retrieved from http://sersc.org/journals/index.php/IJAST/article/view/33586