Feedzai AutoML – A ML Platform for Fraud Prevention that’s 50 Times Faster!

Pranav Dar 07 May, 2019 • 3 min read

Overview

  • Feedzai AutoML is an automated machine learning platform for detecting and fighting fraud in double quick time
  • The developers first used the platform on non-neural network alorithms (XGBoost, LightGBM) to make it easier to work with
  • The tasks it automates are: feature engineering, model training, hyperparameter tuning, and model selection

 

Introduction

Machine learning is THE buzzword in the industry these days as more and more professionals try to jump on the bandwagon. But there’s an obstacle most people can’t seem to cross – getting to grips with the technical nature of the field. Add to that the domain knowledge required, and the majority of aspiring data scientists seem to fall away.

But with the introduction of automated machine learning (AutoML), organizations like Google and IBM are attempting to democratize the field. The aim of AutoML is to lower the barrier for entry in ML so that domain experts can also leverage it, without having to study an entirely new field.

Feedzai is the latest organization to make an entry in this space. They have released Feedzai AutoML, a platform for fighting fraud in double quick time that frees up data scientists from tedious data wrangling tasks. Dealing with risk and fraud cases is a must in the banking industry and while existing data science techniques are adequate, performing the same task at a fraction of the time is a welcome sight.

Feedzai claim that their their AutoML will let data scientist generate one thousand new features (feature engineering) in a matter of minutes. And the most intriguing part? It speeds up fraud prevention workflows by as much as 50 times!

The company realized that businesses were struggling to adopt Google AutoML because of it’s operational difficulty and need for GPUs. So they took a different approach to building their own platform. The developers first used it on non-neural network models like LightGBM and XGBoost so as to speed up the training time. As you can see in the above graph, the model built by 1 machine in 1 day is virtually indistinguishable from a model built by 2 people working for 2 months.

But what truly sets this release apart is the use of an advanced semantic-based automated feature engineering approach. This means that the machine recognizes the semantics associated with each variable in your dataset. The platform can automate the below tasks currently:

  • Automatic feature engineering
  • Automated model training
  • Hyperparameter optimization/tuning
  • Automatic model selection

For a more in-depth explanation of the platform and how to use it for your needs, head over to Feedzai’s blog post.

 

Our take on this

This is quite an interesting release. While I have previously seen AutoML platforms being released, they were not created with a singular task in mind. The results shown by the company are encouraging and promising. AutoML seems to be something that will grow into a big service in the next couple of years.

If you are curious about AutoML and don’t have the money to splurge on learning it, I would recommend checking out Auto-Keras. It’s a recently released open source library that REALLY helps you dig deep into deep learning without getting bogged down by code.

 

Subscribe to AVBytes here to get regular data science, machine learning and AI updates in your inbox!

 

Pranav Dar 07 May 2019

Senior Editor at Analytics Vidhya.Data visualization practitioner who loves reading and delving deeper into the data science and machine learning arts. Always looking for new ways to improve processes using ML and AI.

Frequently Asked Questions

Lorem ipsum dolor sit amet, consectetur adipiscing elit,

Responses From Readers

Clear

  • [tta_listen_btn class="listen"]