arrow left facebook twitter linkedin medium menu play circle
September 20, 2019 - Claire Zhou

How To Accelerate Feature Engineering From Weeks to Minutes

DataVisor introduces Feature Platform, offering out-of-the-box features, feature templates, and complex featuring engineering capabilities to increase coverage and accuracy.

From getting immediate results with out-of-the-box features, to engineering complex features using feature templates in the UI, DataVisor’s new Feature Platform ensures you have the right features for your organizational needs.

Pedro Domingos, author of The Master Algorithm, and a widely recognized authority on AI and machine learning, has been forthright with his opinions on the importance of feature engineering. In an oft-cited quote he poses, and answers, a key question about machine learning features:

“Some machine learning projects succeed and some fail. What makes the difference? Easily the most important factor is the features used.” — Pedro Domingos

As defined in an article by Swetha Basavaraj, Senior Product Manager at DataVisor, “a feature is a characteristic that can help solve a problem using machine learning. The process of extracting such features from a raw dataset is called feature engineering.” 

The Urgent Need for Good Features

To protect against increasingly sophisticated fraud attacks, organizations today are leveraging advanced tools to enhance their fraud protection. However, while machine learning tools and comprehensive rules are widely used to detect modern fraud, their performances vary significantly. 

What makes the difference? As Professor Domingos makes clear, features are the key success factor. This is certainly true when in comes to AI and machine learning-powered fraud prevention.

Feature engineering is an essential part of building any intelligent system. Good features can uncover actionable insights from big data, and transform them for use by machine learning algorithms and rules-based systems. It can be said that features unleash the full power of big data. Without good features, creating high-performing models and rules is a virtual impossibility.

Feature Engineering: Challenges and Pain Points 

Feature engineering is historically complex and time-consuming. Building each new feature is a multi-step process that more often than not requires coding skills. Data scientists spend a significant majority of their time on the data and feature preparation phase before modeling.

An additional challenge for data scientists is that when they want to build complex features and production-ready features, they usually need support from an engineering team. Creating features with complex logic—especially velocity features—requires intensive coding skills. While creating features in the testing environment may be comparatively straightforward, using them in production is another story. Data scientists often need to wait for weeks for engineers to rewrite production-ready code. 

Feature Engineering: 3 Questions Every Organization Must Answer

Given the necessity of feature engineering, and the vital importance of good features, organizations seeking to enhance model performance are confronted with three key questions:

  1. How can we efficiently, effectively, and consistently engineer the most powerful features?
  2. How can we create vast numbers of features in as little time as possible, without sacrificing quality or sabotaging operational efficiency? 
  3. How can our data science teams create high-quality velocity features and production-ready features without having to rely on engineering and IT support?

In an ideal world, there would be a single solution to address all these questions. 

Introducing Feature Platform from DataVisor

DataVisor, the world’s leading provider of AI-powered fraud management solutions, is introducing Feature Platform to fundamentally solve feature engineering pain points and bring exceptional benefits to organizations across industries and use cases. The DataVisor Feature Platform is a critical component of any comprehensive fraud management solution, and can be seamlessly integrated to enhance detection and prevention at scale. With Feature Platform, organizations can accelerate the feature engineering process and empower data science and risk teams to rapidly deploy features to production. Users will leverage global intelligence from more than 4.2B user accounts and draw on advanced deep learning features to successfully increase coverage and reduce false positives.

Create Top-Performing Features Powered by Domain Expertise
With Feature Platform, organizations no longer need to suffer long exploration and testing periods to determine which features will be best for which fraud scenarios. Instead, with Feature Platform, fraud teams can leverage superior domain expertise and global intelligence from 4.2B protected users to automatically get recommendations for top-performing features that will perform best in any given fraud scenario.

Accelerate and Automate Feature Engineering
With Feature Platform, data science and risk teams can significantly speed up the feature engineering process from weeks to minutes. Feature Platform automatically extracts thousands of features from raw data, and enriches the features with AI and machine learning engines.

Examples of Enriched Features: 

  • Enriched IP address features: ip prefix, ip city, check_ip_from_datacenter, and more
  • A feature to check if there is a mismatch between billing address and physical address
  • A feature to check if a first name is part of an email
  • A feature to count the number of application types per day per IP for each branch

Engineer Complex Features and Velocity Features with Built-In Frameworks
Organizations need advanced features and velocity features to detect modern fraud. DataVisor’s Feature Platform is the only platform that provides a wide range of built-in feature frameworks so that users can engineer any complex features with just a few clicks or simple coding, and without requiring additional engineering support.

Example Velocity Features:

  • A feature to calculate the total amount of transactions processed from a particular device where the amount of transaction exceeds $500, within a set 7-day period.
  • A feature to calculate the difference in the price of an item compared to the avg. price of an item across platform.

All features can be directly deployed in production.

Conclusion

Agile and effective feature engineering is critical for modern, AI and machine learning-powered fraud prevention. While there are many options available in the market that can offer support for improved data preparation and model development, only DataVisor’s Feature Platform offers the right combination of speed, production-readiness, and unrivaled domain expertise to power transformative performance.

With the newly-released Feature Platform, organizations can now create advanced fraud features to build sophisticated models and rules with both speed and agility.

about Claire Zhou
Claire is a Senior Product Marketing Manager at DataVisor with over 5 years of marketing experience in security and fin-tech. She is passionate about empowering enterprise customers with AI-based solutions. Her expertise spans data analytics, cybersecurity, and fraud prevention. Claire has an MBA from UCLA.
about Claire Zhou
Claire is a Senior Product Marketing Manager at DataVisor with over 5 years of marketing experience in security and fin-tech. She is passionate about empowering enterprise customers with AI-based solutions. Her expertise spans data analytics, cybersecurity, and fraud prevention. Claire has an MBA from UCLA.