Comments on this page - to the books, papers, conferences etc. - mainly focus on a usefulness of the information involved to the main goal of this project - outlier discovery in multi-dimensional, not necesarily numerical data, exploiting domain knowledge.
Data models that we now have in mind are spatial and spatio-temporal data (e.g. in network traffic, or building management) and relational data (both linked data and multi-relational data in terms of ILP).
Questions for the begining:
What are limits of distance-based methods for those data models?
A distance as defined for first-order logic fomulas, can it help? Which one?
Do we need a formal definition of an outlier?
Are we able to write it?
Outlier detection by a human. Human-assisted outlier detection. What visualization is hepful?
Outlier detection for classified data
Outlier detection and domain knowledge
Integration of Domain Knowledge for Outlier Detection in High Dimensional Space by Sakshi Babbar
Exceptions in spatio-temporal data
Distance-based outliers: algorithms and applications E. M. Knorr, R. T. Ng, V. Tucakov, The VLDB Journal 8, 2000, pp.237-253
only eucledian distances; an attempt to define an outlier formaly as a function of a distance and a number of neighbours;
finding exceptional spatio-temporal trajectories from surveillance
videos: 2D trajectories transformed to aggregates (start/end point,
number of points hggggggyhy (written by a cat; means the length of the
trajectory), heading (ave, max and min of the tanget of the trajectory),
velocity (ave, max, min).
Mining, Indexing, and Querying Historical
Spatiotemporal Data
periodic data, exception movements