Databases Reference
In-Depth Information
Table A.1
VisMiner Modelers
Modeler/Purpose
Advantages
Limitations and Weaknesses
SOM clusterer
Automatically creates subsets of data
observations
Clustering based on Euclidian distance only
Does not provide hierarchical clustering
Adjacent clusters are similar clusters
Clusters generated are non-uniform in size
Can be multidimensional
Nominal data limited to cardinality
30
Clustering may vary, depending on input row
sequence
ANN classifier
Can be trained to fit almost any data
whether linear or curvilinear
Will overfit if not monitored closely
Structure of model difficult to interpret
Readily detects interaction effects
between inputs
No available tests for significance available
May settle in sub-optimal locations
Given random starting location, results may vary
from execution to execution
Nominal data limited to cardinality
10
Decision tree
(classification)
Results are easily visualized and
interpreted
Performance of the model may not be as good as
other classifiers
Fast execution
Nominal data limited to cardinality
30
Support vector machine
(classification)
Can be trained to fit almost any data
Frequently overfits
CPU intensive; with the same dataset, will take
longer than other modelers
Structure of model difficult to interpret
Nominal data limited to cardinality
10
(continued)
