A Mathematical and Computational Framework for Anomaly Detection in Data Streams
Abstract
In this project, we propose to develop a novel, reliable and robust next generation anomaly detection framework for streaming data. Our work will build primarily on our prior research in data streams, and in particular, our recently developed streaming data mining system, Streaming RS-Forest (SRSF). We will enhance SRSF with sophisticated mathematical and computational techniques, and develop new detection algorithms for multiple data streams. To this end, we will pursue the following specific aims. (1) Extend SRSF to the conformal prediction framework so that the prediction of an instance being anomalous can be quantified with confidence. (2) Craft an adaptive incremental version of SRSF by dynamic weighting of current observations and data statistics captured over time. (3) Establish a MapReduce-driven, feature group weighting-based parallelization paradigm for SRSF so that anomalies in high-dimensional streaming data can be efficiently detected. (4) Create a user-driven, adaptively reweighted combination algorithm to further consolidate and enhance the detection generated from distributed SRSFs. (5) Develop a context-aware matrix factorization technique to unveil the underlying (suspicious) patterns from multiple streaming data with missing values.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 11, 2016
- Source ID
- W911NF1510510
Entities
People
- Kun Zhang
Organizations
- Army Contracting Command
- Office of the Secretary of Defense
- Xavier University of Louisiana