Shuvo, Md Sabbir Hossen and Alam, Mohammad Kowshik and Hasan, Md Majedul (2024) Early Risk Detection in Financial Behavior: A Time-Series Study of Transaction Drift and Anomaly Emergence Using Unsupervised Learning. American Journal of Technology Advancement, 1 (12). ISSN 2997-9382

Text AJTA_4190_Early+Risk+Detection.pdf Download (929kB)

Abstract

Financial systems create a lot of transactional data, which comes in streams constantly and documents the customer's behavior and activities. Most traditional systems for financial risk detection are focused on detecting unusual transactions once they have happened, with fewer systems providing the institutions with ways to take proactive approaches to risk detection and intervention. The goal of the study is to establish a time-series unsupervised learning paradigm for financial behavior anomaly emergence monitoring and transaction drift analysis to identify the early risks in financial behaviors. The objective of this study is to detect some signs of instabilities in the behavior that precede severe ones in the finances. Behavioral profiles are created on a weekly rolling window basis based on transaction-level features like spending patterns, transaction frequency, merchant interaction behavior, and financial variability features. These temporal behavioral profiles allow time-series tracking of customers' changing financial behaviors over time. The framework proposed, uses the Isolation Forest algorithm to compute the score of the anomalies in the sequential behavioral windows to detect hidden irregularities and progressive escalation of anomaly. Statistical deviation measures such as variance growth, changes in standard deviation and instability in moving averages are employed to measure the drift and progress of behavioral instability. This study also examines pre-anomaly signals that manifest moderate fluctuations and instabilities signals before important abnormal financial events. The framework is designed for financial monitoring and intelligent risk assessment, enabling proactive monitoring of anomaly trends and evolution of financial behavior. This study shows that financial problems can arise gradually, measured as behavioural drift and not in isolated instances. Experimental expectations show that the anomaly scores get higher and higher when major financial disruptions are about to occur, thereby indicating the possibility of early-stage anomaly emergence detection by applying unsupervised temporal analytics. The proposed framework is an effort to contribute to the financial artificial intelligence research, which is an extension of the integration of rolling-window behavioral profiling, statistical drift measurement and unsupervised anomaly detection into a combined early warning system. This study could be helpful for banks, fintech firms and financial institutions in their efforts to develop fraud prevention, financial stress monitoring, and predictive behavioural risk management in the modern digital financial environment.

Item Type:	Article
Uncontrolled Keywords:	Financial Behavior Analytics, Isolation Forest, Time-Series Analysis, Behavioral Drift Detection, Unsupervised Learning and Financial Anomaly Detection
Subjects:	H Social Sciences
Depositing User:	admin eprints
Date Deposited:	02 Jul 2026 02:42
Last Modified:	02 Jul 2026 02:42
URI:	http://eprints.umsida.ac.id/id/eprint/16718

Actions (login required)

View Item