Using Latent Topics Models to Detect Rare Behaviors

FICO (Fair Isaac Corporation) Computer Science, 2016-17

Liaison(s): Scott Zoldi, Joe Murray
Advisor(s): Robert Keller
Students(s): Savannah Baron, Sneha Deo (PM-S), Emily First (PM-F), Hope Yu

Our project’s goal was to investigate the detection of rare customer behaviors in transactional data using latent topic models, a form of unsupervised machine learning typically used to detect topics in examples of natural language. Our team has developed techniques to apply these models to time series data and has assessed their viability in the detection of anomalous behaviors.