Invalid ad traffic prediction

INVALID AD TRAFFIC
PREDICTION

Real-time prediction of sophisticated invalid traffic to
improve industry rank.

IndustryAdTech

Year2015 – 2018

RoleMachine Learning, Predictive Analytics

TechnologyApache Spark, Scala, HDFS, Apache Zeppelin

Protect your advertisers from fraud

An AdTech player specializing in Programmatic Advertising Platform wanted to pull up it’s ranking in invalid traffic detection during industry surveys. Their requirement was to detect and block sophisticated invalid traffic arriving at the ad-exchange.

AI / ML help spot anomalies in ad traffic

We started off by building a data scraper that collates data from various sources for analysis. We also created a Machine Learning algorithm that predicts invalid ad requests. Moreover, storage optimization techniques on raw HDFS were kicked in to optimize filtered aggregation Spark SQL execution time to <10 seconds for over 9 million records.

IMPACT

Be on top of your game in weeding out IVT

The client was able to improve their Pixalate rank and optimize bid decisions on ad-exchange. The optimized data ingestion pipeline could crunch a TB of data under 50 minutes.