Multiple Kernel Learning based Heterogeneous Algorithm (MKAD)

Related Research Areas
Data Mining and Knowledge Discovery
Project Description
We conducted a study on developing an anomaly detector that can run on heterogeneous data sets. This research resulted in a newly developed version of classical One Class SVMs called MKAD (Multiple Kernel Anomaly Detection) algorithm which can efficiently handle heterogeneous data and conduct fleet wide analysis. We have demonstrated the automated anomaly detection in an off-line mode on large heterogeneous data sets from multiple aircraft. We also demonstrate ability to perform anomaly detection on a data set containing both discrete symbols and continuous data streams and show a 100% detection rate.
Project Administrator(s):
Bryan Matthews,
Santanu Das

Members