Bankruptcy Prediction Model using Machine Learning

Studying financial distress is a crucial component of financial analysis. Financial distress encompasses the difficulties related to cash flow, such as insolvency, default, and business failure or bankruptcy. Bankruptcy is a boundary line that distinguishes a failed company from the rest. This kind of business failure hinders a company’s operations. Thus, it is critical to establish a reliable failure prediction model that can help companies to foresee and survive bankruptcies.

Inaccurate bankruptcy forecasting can have a negative impact. It can cause a catastrophic blow to business owners, partners, society, and the entire national economy. Thus, it becomes the centre of interest for the companies to have a bankruptcy prediction model that can help in decision-making. Technological advancements in the field of economics and computer science have helped in establishing robust machine learning models that can analyze the balanced and imbalanced types of data.

In this article, I am going to analyze and create a failure prediction model using the dataset of bankrupt Taiwanese companies. I will be using F-distribution to analyze the dataset.

An F-distribution determines whether two independent samples have been drawn for the normal populations with the same variance, or if two independent estimates of the population variance are homogeneous or not since it is often desirable to compare two variances rather than two averages.

For analysis, I used sophisticated Python libraries. The accuracy of the bankruptcy model came out to be approximately 98.905%, which indicates the efficiency with which it can forecast failure.

After that, I created an F-distribution table from which I was able to deduce the F1 score of the bankruptcy model. An F-1 score is a weighted average of precision and recall, where an F1 score reaches its best value at 1 and worst score at 0.

This figure shows the f1-scores for the various parameters. Looking at this table, we see that the accuracy of this model is 0.99, which is very close to 1. The weighted average is also coming out to be 0.98. Hence, these figures justify our claim that it is a fairly efficient model.

The third step in the analysis was to create a confusion matrix. A Confusion matrix is an N x N matrix used for evaluating the performance of a classification model, where N is the number of target classes. This matrix consists of 4 elements -- True Positive, True Negative, False Positive and False Negative. In our confusion matrix, the true positive comes out to be 6599, whereas the false negative is 73. It means that 6599 positive class data points were correctly classified by the model and 73 class data points were incorrectly classified as belonging to the negative class by the model.

Thus, this fact has been established that it is now possible to easily predict the business failure that can retard a company’s operations in the future using sophisticated machine learning tools. With the progress made in machine learning, it has now become possible to detect fraud, automate trading activities, and provide financial advisory services to investors. It can analyze a huge amount of datasets within a short time to improve the outcomes without being explicitly programmed. Machine learning is also being widely used in cryptocurrencies and blockchain technologies. With such advances being made in this field, a day will surely come when we will be able to limit the losses being incurred to firms and governments due to business failures by accurately predicting and finding solutions for them using machine learning.

---

References:-

• https://towardsdatascience.com/machine-learning-in-finance-why-what-how-d524a2357b56

• https://scikit-learn.org/stable/modules/generated/sklearn.metrics.f1_score.html

• https://www.analyticsvidhya.com/blog/2020/04/confusion-matrix-machine-learning/

• https://corporatefinanceinstitute.com/resources/knowledge/other/machine-learning-in-finance/#:~:text=In%20finance%2C%20machine%20learning%20algorithms,outcomes%20without%20being%20explicitly%20programmed.

• https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0254030

• https://archive.ics.uci.edu/ml/datasets/Taiwanese+Bankruptcy+Prediction