Financial Fraud Prevention using AI

Real-time transaction approval process that improves speed and stability of approving transactions by using named entity restrictions based on machine learning models. The process involves identifying entities involved in transactions from past data, creating rules specifying authorized entities, and iteratively training machine learning models to accurately identify entities in real-time. The identified entities are compared against the rules to approve transactions. This allows faster approval compared to manual review while still enforcing entity restrictions.

Source

Generating a virtual asset wallet address blacklist using graph attention networks (GAT) to proactively detect high-risk virtual asset wallets that are prone to illegality. The blacklist is generated by training a GAT model using full node indices, common transaction data, and an existing blacklist. The model calculates scores for wallet addresses based on their connections and transactions. Addresses with high scores are added to the GAT blacklist.

Source

Real-time monitoring and prevention of fraudulent transactions in metaverses using parallel metaverses. When a suspicious transaction is detected in a metaverse, a parallel metaverse is created. The original transaction is analyzed in the parallel metaverse to scrutinize it further. Encrypted NFTs representing assets are transferred between users in both metaverses. If the parallel metaverse confirms the transaction, it proceeds in the original metaverse. If fraud is suspected, the original metaverse can freeze activities and temporarily deactivate financial institutions.

Source

Sharing transaction-related data between entities of a payment network using a common transaction ID. The method involves generating a confidence score indicating the likelihood that two transactions are the same using a machine learning model. If the confidence score exceeds a threshold, a common transaction ID is generated based on the attributes of one transaction. This common ID is then used to request additional transaction data from the other entity referencing the common ID.

Source

Detecting potential fraudulent use of identification items like credit cards by estimating if multiple events using cards with identical partial codes are from the same card or different cards based on the time elapsed between events. This allows detecting fraud without exposing full card numbers. It's unlikely for multiple cards with matching partials to be used in quick succession, so a short time gap indicates potential fraud.

Source

Automated system to detect and freeze cryptocurrency wallets used for laundering illicit funds. The system traces cryptocurrency transactions through mixing organizations to identify all their wallets. It then initiates secondary transfers between known wallets using the mixing wallets. By monitoring these transfers, it can detect patterns indicative of obfuscating illicit funds transfers. Suspect wallets are then frozen.

Source

Fraud detection in service provider systems using machine learning models. The system involves using two machine learning models to improve fraud detection. The first model is a transaction-level model that determines fraud probability based on transaction features. The second model is a user-level model that determines user fraud probability based on user behavior patterns. The system combines the outputs of both models to make a final fraud decision. If the transaction-level model indicates high fraud probability but the user-level model indicates low fraud probability, it is a potential false positive. Similarly, if the transaction-level model indicates low fraud probability but the user-level model indicates high fraud probability, it is a potential false negative. In such cases, the system performs additional checks or escalates the transaction for manual review to mitigate errors.

Source

Using machine learning to identify combinations of attributes that cause abnormal behavior in devices or payment transactions. The approach involves training a machine learning model on historical data to correlate combinations of attribute values with abnormal outcomes. Then, for new devices or transactions, if the model indicates a high rate of abnormality for their attribute combination, it can flag them as potential issues. This allows faster and more scalable detection of problematic attribute combinations compared to brute force scanning.

Source

Detecting suspicious payment cards at point-of-sale terminals and ATMs to prevent card cloning fraud. The technique involves capturing images of inserted payment cards and analyzing them to determine if they are authentic or cloned. The images are compared against known characteristics of genuine cards to detect deviations. Factors like bank logos, card designs, and wear patterns are analyzed. If a card is flagged as suspicious, an alert is generated and the transaction can be blocked.

Source

System for detecting money laundering through gaming machines by monitoring and analyzing machine usage patterns. The system uses machine learning models to identify potential money laundering activity by generating inferences from learned patterns. It monitors gaming machines for behaviors indicative of money laundering and notifies the casino if likely laundering is detected. The models analyze factors like win/loss ratios, betting patterns, and transaction volumes to identify suspicious activity.

Source

Counter fraud system that uses device signatures to proactively detect and prevent fraud. The system compares a device's identifying parameters to signatures of devices associated with fraud. If similarities are found, it raises suspicion and can initiate security measures like 2FA or transaction denial. The device signatures are built by analyzing parameters of devices used in fraud.

Source

Monitoring electronic transactions to detect potential fraud by comparing the location of the user's device to the transaction location. When a transaction is initiated, the device location is checked against the last known location from cached data. If they don't match, an alert is generated. This leverages the fact that a user's device usually stays nearby during a transaction. If the device is far away, it could indicate fraud.

Source

Dynamic fraud detection system that prioritizes important fraud events based on factors like fraud type, activity, and temporal information. The system uses a fraud importance engine to calculate scores for fraud events that are fed into the fraud detection engine. This allows the detection engine to prioritize important frauds and optimize performance metrics. The fraud importance scores are generated from user-defined parameters like fraud type, cross-channel activity, etc.

Source

Verification of unstructured data like invoices using optical character recognition (OCR) and database lookup to accurately extract and analyze data elements like account numbers, company registrations, and signatures. The method involves identifying predetermined data elements in unstructured sources like invoices using OCR, then using those elements to access a database and retrieve verified data. This verified data can then be used for further analysis of the unstructured data. The system can also check for fake invoices by analyzing signatures and cross-referencing with a database.

Source

Reducing resource consumption and latency of fraud detection systems in commerce platforms with high transaction volumes. The method involves leveraging feature extraction and machine learning techniques to optimize fraud detection. Instead of generating and communicating all features for each transaction, a subset of critical features are identified and precomputed offline. These precomputed features are then cached and reused for similar transactions, reducing the resource-intensive feature generation and communication. The remaining features are dynamically computed on-the-fly. This partial feature reuse and dynamic computation reduces resource consumption and latency compared to generating all features for each transaction. The precomputed features are based on machine learning models that identify critical feature combinations for efficient fraud detection.

Source

Detecting and blocking card testing attacks (CTAs) in financial transactions without introducing excessive latency. The technique involves asynchronously scoring transactions for CTA risk using separate ML models and storing the CTA scores in a cache. When a transaction comes in, the stored CTA scores are retrieved and used to adjust the blocking score thresholds. The transaction is then evaluated against the adjusted thresholds to determine if it's allowed or blocked. This allows efficient CTA detection without adding latency to the charge path.

Source

Early detection of first party fraud in financial transactions by predicting future fraud instances. The method involves aggregating normalized attributes of multiple activities associated with a user from different sources over time, then feeding those normalized attributes into a machine learning model trained to output the likelihood of a future fraud instance for that user based on normalized attributes of past fraud instances. If the predicted likelihood exceeds a threshold, an alert is sent. This allows detecting potential fraud schemes before they become apparent by identifying users with elevated fraud risk based on patterns in their activity.

Source

Using machine learning to analyze past transactions to optimize network efficiency, enhance approval rates, mitigate fraud, and enrich customer profiles. The method involves converting groups of transactions into sequences of transaction features, generating tokens based on those sequences, feeding the tokens to an ML model to estimate future transactions, and using those estimates to approve new transactions. The ML model learns to predict transactions based on historical patterns.

Source

Using graph neural networks to analyze transaction and external data to identify merchants with negative settlement risk in electronic payment processing. The method involves generating a graph data structure with nodes representing merchants, transactions, etc., and edges connecting related nodes. Text representations of transaction and external data are concatenated to generate node embeddings. Risk scores are propagated through the graph to identify merchants with negative settlement risk. This allows proactive risk mitigation and transaction authorization based on graph analysis rather than just transaction parameters.

Source

Proactive fraud prevention system for payment cards that allows users to define and control rules for their cards. The system automatically creates rules based on transaction history and demographics, allows users to view and modify the rules, and enforces them to allow or restrict payments. It sends alerts when rules are triggered and updates rules in real-time based on user input.

Source