Click Fraud Detection: HK-Index for Feature Extraction from Variable-Length Time Series of User Behavior

Abstract

Most conventional time series outlier detection techniques exhibit limitations for variable-length time series. To tackle this problem, we propose HK-index to extract features from variable-length time series for any outlier detectors. HK-index identifies the longest subseries which satisfies the conditions by a pre-defined parameter <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$k$</tex> . The length of the longest subseries found is then used as the feature. The novelty of core idea of HK-index is to use the information of subseries to present the whole time series, ignoring the irrelevant information. HK-index has linear time complexity and ease of application. It has been utilized in Alibaba to filter the daily fraud traffics.