Page 153 -

140 4 Statistical Classification

This decision tree provides useful insight concerning individual contributions to
the target classification, especially when accompanied by an easy to read graphical
display. In the present case, the clinician is able to appreciate that only two of the
four variability features are required to distinguish FS from non-FS situations.
Feature ALTV (abnormal long-term variability) is the one that contributes most to
this distinction with a low percentage of false positives (5 in 2057). The overall
classification error is achieved with remarkably high sensitivity 92.8 % (64169) and
specificity 99.6 % (204912057).
The decision tree design and results are obtained very quickly (Statistics
processing time about 1 second on a 733 MHz Pentium) making this approach
adequate for data mining applications.
In a more realistic data mining application in the same medical area, one would
want to apply this tool for the determination of the binary tree that best
discriminates the three major classes of foetal well-being, normal, suspect and
pathologic, based on foetal heart rate features. Again, this can be achieved in an
effective and efficient way with the CART approach (see Exercise 4.26).

Bibliography

Argentiero P, Chin R, Baudet P (1982) An Automated Approach to the Design of Decision
Tree Classifiers. IEEE Tr Patt An Mach Intel, 451-57.
Batchelor BG (1974) Practical Approach to Pattern Classification. Plenum Pub. Co., New
York.
Bell DA (1978) Decision Trees, Tables and Lattices. In: Batchelor BG (ed) Pattern
Recognition. Ideas in Practice. Plenum Press. New York, pp. 119-141.
Breiman L, Friedman JH, Olshen RA, Stone CJ (1993) Classification and Regression Trees.
Chapman & Hall I CRC.
Centor RM (1991) Signal Detectability: The Use of ROC Curves and Their Analyses.
Medical Decision Making, 1 1 : 102-106.
Chang CY (1973) Dynamic Programming as Applied to Feature Subset Selection in a
Pattern Recognition System. IEEE Tr Syst Man and Cybern 3:166-171.
Chen CH (1982) Application of Pattern Recognition to Seismic Wave Interpretation. In: Fu
KS (ed) Applications of Pattern Recognition. CRC Press Inc., Boca Raton, pp. 107-120.
Cherkassky V, Mulier F (1998) Learning from Data. John Wiley & Sons, Inc.
Chien YT (1978) Interactive Pattern Recognition. Marcel Dekker Inc., New York.
Dattatreya GR, Sarrna VV (1981) Bayesian and Decision Tree Approaches for Pattern
Recognition Including Feature Measurement Costs. lEEE Tr Patt An Mach Intel 3:293-
298.
Devijver PA (1982) Statistical Pattern Recognition. In: Fu KS (ed) Applications of Pattern
Recognition, CRC Press Inc., Boca Raton, pp. 15-35.
Devijver PA, Kittler J (1980) On the Edited Nearest Neighbor Rule. Proc. 51h Int. Conf, on
Pattern Recognition, 72-80.
Dubuisson B (1990) Diagnostic et Reconaissance des Formes. ~ditions Hermes, Paris.
Duda R, Hart P (1973) Pattern classification and scene analysis. Wiley, New York.