Case Study: Fraud Detection Through User Behavior Analytics in Financial Services

Introduction:

In this case study, we dive into the patterns of suspicious user activity to better understand and detect potential fraud risks. By analyzing key performance indicators (KPIs) such as transaction amount, session duration, and failed login attempts, we uncover trends, pinpoint regional hotspots, and identify device-specific risks to improve monitoring strategies.

Summary of Results

1. Seasonal Patterns in Suspicious Activity: Suspicious activity in terms of transaction amounts and session durations revealed a seasonal trend, with declines in February 2022 and spikes in early 2023 and 2024. Failed login attempts, which remained relatively low in 2022, rose significantly by 2024. November 2024 data shows a drop, likely due to incomplete data for that month.
2. Geographic Insights: Maps highlighted that Lagos exhibited lower levels of suspicious activity across all fraud detection KPIs, while Abuja showed elevated activity, especially in areas beyond failed login attempts. This suggests a need for intensified monitoring in Abuja.
3. Withdrawal Risks: Analysis of user behavior during transactions showed that withdrawals are particularly prone to suspicious activity, aside from login attempts. This underscores the importance of close monitoring during withdrawals and fund transfers to prevent fraud.
4. Device-Specific Trends: Both mobile and desktop devices saw high levels of suspicious activity, with mobile devices displaying elevated risks in terms of transaction amount and session duration KPIs. Monitoring both platforms is essential for comprehensive risk management.

KPIs and Thresholds

1. Suspicious Users by Transaction Amount: Transactions exceeding ₦20,000 are flagged as suspicious.
2. Suspicious Users by Session Duration: Session durations shorter than 2 minutes are considered suspicious.
3. Suspicious Users by Failed Logins: Users with more than 3 failed login attempts are marked as suspicious.

These thresholds can be adjusted based on specific requirements and risk tolerance.

Dashboard Analysis

Dashboard 1: Suspicious Users Over Time

This dashboard reveals a clear seasonal pattern in suspicious transactions. February 2022 experienced a drop in suspicious activity, which then rose in January 2023. A similar rise was seen in January 2024 for session durations, showing that suspicious activity often peaks at the beginning of the year. Notably, 2024 saw an increase in failed login attempts. The sharp declines in November 2024 are likely due to partial data.

Dashboard 2: Suspicious Users by Location

Geographic analysis maps show that Lagos experiences relatively low suspicious activity across all KPIs, while Abuja shows higher activity, particularly beyond failed login attempts. This points to a greater need for targeted fraud detection and monitoring in Abuja to reduce risks.

Dashboard 3: Suspicious Users by Transaction Type

The charts illustrate that suspicious activity is often associated with withdrawals, excluding failed login attempts. This finding highlights the need for enhanced security during withdrawal and transfer processes to detect potential fraud more effectively.

Dashboard 4: Suspicious Users by Device Type

Analysis across devices shows a higher prevalence of suspicious activity on mobile and desktop, particularly for KPIs like transaction amount and session duration. Mobile devices exhibit notably higher suspicious behavior, emphasizing the need for stringent monitoring on both mobile and desktop platforms.

Conclusion

This case study underscores the importance of tracking seasonal patterns, regional activity differences, transaction types, and device-specific behavior to detect potential fraud risks effectively. Through targeted monitoring and threshold adjustments, organizations can better safeguard against suspicious activities and potential fraud. Future studies may explore adapting these KPIs and thresholds to specific needs, further refining fraud detection capabilities.

If you’d like to get the Tableau workbook for a more detailed look, contact us at info@kcinnovix.com