Non-Obvious Root Cause Analysis
Publication Date: 2015-Feb-09
The IP.com Prior Art Database
The features deals with methods and systems to extract the non-obvious root cause analysis (or likeliness behavior) of entities chosen by a user and its Implementation on Visualization and BI systems via Metadata Enhancement.
Page 01 of 4
Non-Obvious Root Cause Analysis
Problem Statement and Existing Solutions:
Often it is felt to conduct a root cause analysis on some entities of business value as they are seen on a business intelligence or visualization system. At times these entities might not have an obvious relation among them in the current layout.
Sometimes even while changing the facts/ dimensions of the visualization or BI also no apparent relation can be discovered across any combination of the dimension/ facts which can describe the root cause of co-existence/ like behavior of the entities of interest.
At such occasions the need of a method is felt which can do some complex modeling to extract the non-obvious root cause of the given behavior (likeliness)
The features deals to extract the non-obvious root cause analysis (or likeliness behavior) of entities chosen by a user and its Implementation on Visualization and BI systems via Metadata Enhancement.
Phase - I: Optimal Separation Space/ Axes Extraction:
Updating the metadata with identification of the Selection Set (Ss) and Non Selection Set (So)
Rotating the Selection Set orthogonally and extracting the top 3 (or lesser in case of less than 3 dimensions/ facts in the data) to form the Principal Component Matrix for the Selection Set (Ӣեp-s) with dimensions (1,n)
Rotating the Non-Selection Set orthogonally and extracting the top 3 (or lesser in case of less than 3 dimensions/ facts in the data) to form the Principal Component Matrix for the Non-Selection Set (Ӣեp-o) with dimension (m,1)
Obtaining the correlation matrix (Ӣեc) across the Component matrices Ӣեp-s and Ӣեp-o with dimension (m,n)
Obtaining the Transpose (Ƭc) matrix of the correlation matrix Ӣեc
Obtaining the Optimal Space Matrix (Ӣեos) with dimension (m,1) by the multiplication of the Correlation Transpose Matrix Ƭc and the Selection Set Component Matrix Ӣեp-s.
Selection of the two (or n Axes) axes with the highest scalar components of the Optimal
Page 02 of 4
Space Matrix for a two (or n) dimensional layout.
Loading the Optimal Axes data to the the existing dataset as additional facts.
Phase - II: Constrained Centroid Space Clustering in the Optimal Separation Plane:
Identification of the space for the play of the constrained centroid by extending the boundaries of the peripheral points of the Selection set by an area/ volume etc. (circle or sphere etc . depending upon the dimensions of the optimal space) with radius 0.5 times the Euclidean distance between the peripheral point/ entity and the entity closest to it.
Identifying the Initial positioned of the constrained space centroid by considering the centroid of the Selection Set entities, assigning equal weights to each entity.
Iteratively choosing 2…N nos. of centroids at random positions outside the selection
o For each of the above iteration progressing and completing the clustering proces...