arrow left facebook twitter linkedin medium menu play circle

Digital Fraud Wiki

Your source for the latest fraud intelligence, insights, research, and commentary.

Feature Engineering

Feature engineering is the process of using domain knowledge to extract features from raw data via data mining techniques. High-quality features are important for fraud detection, because a machine learning model is only as good as the data and features it is fed. Today, data scientists spend 70% of their time on data cleaning and feature engineering. DataVisor makes the feature engineering process a lot easier by offering an advanced and automated Feature Platform.

Why Is DataVisor’s Feature Platform Unique? 

Out-of-the-Box Feature Packages for Fraud Use Cases

DataVisor’s Feature Platform provides pre-built feature packages that are optimized for specific fraud types, including application fraud, ATO, transaction fraud and other scenarios. Extensive out-of-the-box features based on DataVisor’s domain expertise reduce time spent manually building features from scratch.

Deep Learning Features and the Global Intelligence Network

Deep learning helps detect script-generated content, disposable email services, anonymous proxies, suspicious IP usage and more. Gather intelligence from DataVisor’s consortium database to provide unique signals such as entity ages, geo-usage profiling and blacklists.

Backtesting on Historical Data and Rapid Deployment

Easily backtest on extensive historical data to validate detection performance, with full flexibility to choose the data’s timestamp and sample set. Results are returned within seconds; all the features are production-ready and can be deployed in minutes.

High Scalability and Big Data Architecture

Handle years of data for large institutions and process massive amounts of data in real time — 10,000+ QPS with only 10-50 ms latency. To minimize computation costs, DataVisor builds features on top of features, so that each feature is computed only once.