Fraud detector endpoint for online advertising, submitted to Developer Week 2015 for the Virool Challenge
The rules for the challenge are here:
http://devweek.virool.com/rules
We first extracted features using iPython, and generated test .CSV files to explore the data in Tableau
Once we had a set of about 15 features, we proceeded to create a Support Vector Machine model using Python scikit-learn, with a training set consisting on 80% of the data set, randomly selected.
To set up an endpoint capable of handling 1000 request per minute, we set up 16 Tornado processes with an Nginx load balancer upfront, using an Azure Ubuntu 8-core server
http://wwwhisper.cloudapp.net?ip=xxxx&user_agent=yyy&referer=zzz