Early Prediction of Banking Closure Failures Using Machine Learning for Anomaly Detection

Abstract

Closing the banking system is one of the most crucial things that should be done in financial
institutions especially banks. This can happen daily, monthly, or yearly. When the bank follows a
proper routine every day, it helps maintain consistency in its operations, and it will also help
maintain the internal and external balance of the bank. The closing is stopped and delayed when any
mistake or unusual and suspicious transaction occurs. Thus, it adds stress, work, and effort. This
raises the dangers of doing business. Banks must close every day to deal with the large number of
transactions that happen during the day that the bank’s actual activities to directly find any
mistakes, suspicions, fraud attempts or theft. Unlike the shops or stores that are mostly closed for a
day in the month or a year, this is different. The research proposes a Machine Learning based
approach for the early identification of anomalies leading to daily closing failure. We used synthetic
data that acted like a real bank setting. The dataset used was open-source and available on Kaggle
as well. They used several models such as Random Forest, which is a supervised model, and
Isolation Forest, which is an unsupervised model The models were deployed and analyzed on
performance through metrics pertaining to machine learning such as recall, positive accuracy, F1-
Score, and ROC-AUC. The study showed the results of Random Forest model are highly accurate
for classifies data. The model called Isolation Forest was also quite capable of finding the
unexpected patterns. The applications of both models together could be beneficial in developing and
constructing a model for an early warning system which would reduce sudden closures and enhance
the role of digital transformation in the banking industry as a whole