Tools for detecting financial fraud and fine-tuning fraud detection AI algorithms.

Problem Context

Detect Money Laundering, Unauthorized Transaction, Embezzlement, and Third Person Fraud Schemes

EVA (Event Detection with Visual Analytics), 2016, was created in collaboration with Erste Bank to tackle financial fraud detection (FFD) challenges faced by institutions handling large volumes of transactions. Traditional AI and data mining techniques were insufficient alone, thus necessitating a tool designed to integrate of Visual Analytics of the complex data. EVA aims to enhance the detection and prediction accuracy in FFD systems, providing support for fraud investigation and fine-tuning AI detection algorithms' performance.
See EVA full article .pdf

NEVA (Network EVA), 2019, addresses the complexity of detecting fraudulent networks in financial transactions. Traditional fraud detection methods, focused on individual behaviors, fail to capture complex schemes like money laundering or straw person frauds. NEVA integrates visual analytics to explore these networks, aiming to reduce false negatives and positives in fraud detection.
See NEVA full article .pdf

Defining Success

Qualitative studies about the effectiveness of the approach, as interpreted through Klein's model.

Success in EVA's and NEVA's context was centered on designing and developing a system that efficiently identifies fraudulent transactions, minimizes false positives, and adapts to changing patterns. The goal was to create a Visual Analytics tool that enhances fraud detection by integrating with existing AI systems and offering superior insight for investigators.

In evaluating this system, we performed usability sessions and analyzed qualitative data (notes, audio, video) using Klein's model for its broad intelligence analysis. This analysis was categorized into Connection, Coincidence, and Curiosity insights, focusing on integrating multiple data views, uncovering hidden relations, and sparking interest in data anomalies. This method offered a comprehensive way to assess the tool's effectiveness in exploration and sensemaking in data analysis.

Main methodology reference:
"Seeing What Others Don't" - Gary Klein


Making sense of multidimensional time-oriented data

EVA's design, focusing on usability and interaction, employs visualization techniques and automatic fraud detection methods in financial transactions. Collaboratively developed with domain experts, it features multiple views for data analysis, including temporal and score construction views, scatterplots, and dynamic tables. These interconnected views support domain experts in fraud analysis, emphasizing practical utility and domain-specific needs in its visual and interactive design.

For a more interactive and detailed understanding of EVA's capabilities and design, I highly encourage you to watch the use case video provided alongside this document. The video offers a dynamic and practical demonstration of how EVA operates in real-world scenarios, enhancing your grasp of its features and functionalities in action.

NEVA's design, focusing on the analysis and detection of fraudulent networks in financial transactions, integrates visual analytics with advanced detection methods. Developed in close collaboration with domain experts, NEVA offers a multifaceted environment for exploring complex data relationships. Its features include multiple coordinated views for comprehensive data analysis, a guidance-enriched component for network pattern generation, detection, and filtering, and an interactive exploration environment. These elements work together to support the investigation of customer networks, aiming to reduce both false-negative and false-positive alarms in fraud detection. NEVA's design emphasizes practical utility and meets domain-specific requirements through its visual and interactive capabilities.

For a more interactive and detailed understanding of NEVA's capabilities and design, watching the accompanying video is highly recommended. The video provides a dynamic and practical demonstration of NEVA in action, enhancing your understanding of its features and functionalities.

For a quick and simplified understanding of how the AI scoring algorithms function, watch this concise 53-second video clip.


Participants found EVA & NEVA intuitive and more dynamic compared to traditional tools
4 Years
of iterative design

11 FFD
Expert Investigators

6 Articles

The evaluation results of EVA highlighted its effectiveness in providing intuitive and dynamic visual analytics, particularly in the context of fraud detection. Participants successfully navigated and completed tasks, deriving valuable insights mainly of the 'connection' type, which involved uncovering new information by linking events or variables within the complex data.

The evaluation of EVA highlighted its potential evolution into NEVA (Network EVA), integrating network analysis to enhance detection of complex fraud schemes by comprehensively mapping bank account relationships. This progression represents a significant step forward in the realm of visual analytics, enhancing the ability to uncover and understand complex fraud patterns within the financial sector.

As consequence, NEVA was designed, developed and evaluated. The NEVA system demonstrated remarkable results in enhancing the detection and analysis of complex fraudulent networks in financial transactions. The added value of NEVA is suppported by (1) the analysis of insights gained with the help of NEVA and (2) the additional fraudulent cases identified with the help of NEVA that were later confirmed by investigators. NEVA confirmed its usability and efficiency.

CV & Contact

Feel free to get in touch!

You can contact me via: