A Galaxy of Correlations - Detecting Linear Correlated Clusters Through K-Tuples Sampling Using Parameter Space Transform

MCML Authors

Daniyal Kazempour

Dr.

* Former Member

→ Group Thomas Seidl
 Database Systems, Data Mining and AI

Peer Kröger

Prof. Dr.

Principal Investigator

* Former Principal Investigator

Thomas Seidl

Prof. Dr.

Director

Database Systems, Data Mining and AI

Abstract

In different research domains conducted experiments aim for the detection of (hyper)linear correlations among multiple features within a given data set. For this purpose methods exist where one among them is highly robust against noise and detects linear correlated clusters regardless of any locality assumption. This method is based on parameter space transformation. The currently available parameter transform based algorithms detect the clusters scanning explicitly for intersections of functions in parameter space. This approach comes with drawbacks. It is difficult to analyze aspects going beyond the sole intersection of functions, such as e.g. the area around the intersections and further it is computationally expensive. The work in progress method we provide here overcomes the mentioned drawbacks by sampling d-dimensional tuples in data space, generating a (hyper)plane and representing this plane as a single point in parameter space. By this approach we no longer scan for intersection points of functions in parameter space but for dense regions of such parameter vectors. By this approach in future work well established clustering algorithms can be applied in parameter space to detect e.g. dense regions, modes or hierarchies of linear correlations in parameter space.

inproceedings

---

EDBT 2019

22nd International Conference on Extending Database Technology. Lisbon, Portugal, Mar 26-29, 2019.

---

Authors

D. Kazempour • L. Krombholz • P. Kröger • T. Seidl

---

Links

PDF

---

Research Area

 A3 | Computational Models