Filtering the noise from an avalanche of Google Analytics Metrics : Anomaly Detection
At Tatvic, we have built an Anomaly Detection Engine that alerts the analyst about sporadic changes in Google Analytics metrics. Additionally, the analyst can also drill down into the possible root causes of the anomaly which enables him to take quicker business decisions.
This talk will focus on the methods used for Anomaly Detection as well as its utility to the end user.
Analysis and drill down used to be a simpler problem earlier. Plot each metric on its dashboard and keep updating it regularly. As Google Analytics has vastly increased its coverage of metrics, having a dashboard for each metric does not solve the problem. The analyst might easily miss out on some of the key changes in their data. Imagine the loss caused to an eCommerce website if the Page Load Time of their Home Page spikes. Anomaly Detection Systems can aid the analyst figuring out these sporadic patterns faster and taking quicker business actions. Once an anomalous metric is discovered, it is possible to drill down into why that happened.
The challenge herein lies in detecting anomaly patterns with a fair amount of accuracy as well as transforming it into insights that are immediately useful. Some of the open source technologies used are the Python Scientific Stack (Algorithms), Mongodb (Backend) and d3.js (Frontend)
Kushan Shah is a Web Analyst at Tatvic. He works towards solving business problems using a combination of data and algorithms.