Unsupervised outlier detection systems for e-commerce

Published

October 16, 2018

Abstract

For large e-commerce businesses, monitoring the rate at which users convert on their websites is essential. Given conversion rate data exhibits certain unique characteristics such as sub-daily time series, multiple seasonality, presence of trend and strictly non-negative values, detecting outliers by business users is non-trivial. We have explored a range of time series outlier detection methods to test and build a system for automatic outlier detection for e-commerce businesses.