Unsupervised shallow learning for fraud detection on marketplaces
Adyen provides payments processing and financial services to many marketplaces like eBay, GoFundMe or Wix among many others. In this setting, any individual can sign up and start selling or providing services and thus there is a need for strong requirements around behaviour monitoring to prevent illegal or damaging situations.
This talk will take you through our journey when solving for behaviour prediction and monitoring for an ever-growing dataset. A journey based on several iterations first on posing the problem, having limited or non-existing labels, and then on the different mathematical and technological implementations leading to a solution that leverages an open source stack (airflow, spark, keras and tensorflow) over python.
We will not only cover the solutions we arrived at with code, formulas and memes, we will also showcase results and lessons learned over this journey such as:
- mathematical insights for machine learning models
- distributing deep learning over spark
- user experience around machine learning
- model tracking
- validation of unsupervised algorithms.
Andreu works at Adyen as VP of data science and ML, coordinating all the data science and machine learning efforts around the company and providing technical guidance to teams and individuals. His previous role at Adyen was data scientist, tech lead and individual contributor in the area of predictive monitoring and network risk. Before that, Andreu worked for 12 years in the aerospace industry, both at private companies and at the European Space Agency, in the area of data mission analysis and algorithmic processing. Andreu holds a MSc in Telecomm engineering by the Universitat Politecnica de Catalunya in Barcelona (UPC).
visit the speaker at: Homepage