aceAI and NLP models for detecting online fraud from text data
Source article: Applications of AI-Based Models for Online Fraud Detection and Analysis
Abstract Background Fraud is a prevalent offence that extends beyond financial loss, impacting victims emotionally, psychologically, and physically. Advances in online communication technologies continue to create new opportunities for fraud, and fraudsters increasingly using these channels for deception. With the progression of technologies like Generative Artificial Intelligence (GenAI), there is a growing concern that fraud will increase in scale using these advanced methods, with offenders employing deep-fak…
Contested: both sides are scored from claims and sources, not community votes.
"Scam-phishing-fraud-email-attack-mail-online-system-cybercrime-information-access-credit-money-hack-hacker-laptop-malware-password-protection-software-steal-text-graphic-design-illustration-Material-property-techno" by Mohamed Hassan is marked with CC0 1.0. To view the terms, visit https://creativecommons.org/publicdomain/zero/1.0/deed.en/.
On June 13 2025, Crime Science published a systematic literature review of AI and NLP for online fraud detection. The authors screened 2457 records and analyzed 223 studies, mapping data sources, algorithms, and evaluation metrics across 16 fraud types and summarizing best-performing methods for detecting scams in text.
The findings matter because fraud causes financial, emotional, and psychological harm and is expected to grow with generative AI and deep-fakes, yet current detection research remains fragmented by scam type. Uncertainty remains about real-world generalizability, durability against evolving tactics, and reliability of reported performance due to omitted limitations and inconsistent reporting.
- Systematic review screened 2457 records, included 223 studies under PRISMA-ScR criteria focused on text data and AI methods.
- Review identified 16 different fraud types studied separately, with data sources, algorithms, and performance metrics varying across studies.
- Authors report best-performing recent AI methods for scam detection but note inconsistent metric reporting across studies.
AI and NLP models trained on text data can detect and analyze patterns across multiple categories of online fraud.
Fraud-detection models trained for specific scam types often fail to generalize to new fraud types and lose effectiveness when trained on outdated data, with inconsistent performance reporting.
The rundown
The review followed PRISMA-ScR, screening 2457 academic records, finding 350 eligible and analyzing 223 that used text data and AI methodologies for online fraud.
It catalogued data sources, algorithms, and evaluation metrics, identifying 16 distinct fraud types studied in isolation and summarizing recent best-performing detection methods.
Conclusions highlight lack of generalizability across fraud types, degradation from outdated training data, omitted discussion of data limitations and training biases, and selective metric reporting that may bias evaluation.
Review finds primary studies often omit data limitations and training biases, use outdated training data that reduces effectiveness as scams evolve, and report performance metrics inconsistently.
Sources
- Peer-reviewedCrime Science2025-06-13
How should this claim be treated?
The debate