SUMMARY
- Best error rate on the scenic estimation problem is 43%
using RGB+LL+ENT
- Features extracted from images in the scenic
estimation problem do not give any recognizable pattern
- Best error rate on the surname generation problem is 39%
using context of 5
- Context plays an important role in determining
the sound but we need to explore new features to improve accuracy
- Implemented Bayesian decision tree
- Developed a framework for decision tree algorithms
- Demonstrated improvements over IND (can handle
large data sets with large number of classes and attributes,
special characters for class and attribute values)
- Software is available from the ISIP web site at:
/hse/ies/resources/technology/software/1998/decision_tree