Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
*
FEUP – Faculdade de Engenharia da Universidade do Porto, Portugal
E-mail: fsereno@fe.up.pt
†
HSJ – Hospital de S. João, Dep. Ginecologia e Obstectrícia, Porto, Portugal
††
FMUP - Faculdade de Medicina da Universidade do Porto, Portugal
INEB - Instituto de Engenharia Biomédica, Porto, Portugal
( y1 , x1 ),K , ( y l , x l ). Q( yi , f ( xi )) = yi − f ( xi ) ε (8)
It can be shown that under particular conditions where the function |.|ε , is called ε-insensitive loss.
there is uniform convergence to the mathematical The function given has the general form:
expectation (3) of an empirical measure estimator,
f ( x ) = ∑ i=1 ci K ( x, x i )
e.g. the functional (4), therefore l
(9)
Page 2
7 Experimental results 8 Conclusions
The SVR training algorithm [8] has been tested on SVR is equivalent to maximizing the margin
two subsets of our Foetal weight (FW) data set, between training examples and the regression
each one corresponding to the inferior and superior function. It is an alternative to other neural
tails of the FW distribution function, as shown in networks with training methods that optimize cost
figures 1 and 2. The central and most frequent functions such as the mean square error, therefore
cases will not belong to our sub-sets. The it can be applied to FW estimation.
experiment consists in the performance
SVR is motivated by the statistical learning theory,
determination in a test separate set.
which characterizes the performance of SVR
The predicted FW (FWpred) was computed from learning using bounds on their ability to predict
two echographic features abdominal circumference future data.
(AC) and femur length (FL), in two different
The training consists in solving a constrained
portions of the distribution function, with almost
quadratic optimization problem [4,5,9]. Among
the same number of examples. The polynomial
others, this implies that there is a unique optimal
kernels used in this experiment were of order ≤ 7. solution for each choice of the SVR parameters.
The ε-insensitive loss function used values of This is unlike other learning machines, such as
ε >0.05. The number of support vectors returned by standard Neural Networks trained using
our algorithm was SVinf=90.5% and SVsup=96.5%, backpropagation.
respectively in the inferior and superior tails of the
FW distribution function.
Finally, the error rates we got were Einf=11.2%
Esup= 10.0%, in the inferior and superior tails of the
FW distribution function, respectively.
4000
4000
FW (blue) - Estimated FW (red)
3500
3500
3000
3000
2500
2500
2000
2000
1500
1500
1000
5 10 15 20 25 30 1000
#case 5 10 15 20 25 30 35
#case
Figure 1 – Support Vector Regression (SVR) Figure 2 – Support Vector Regression (SVR)
predicted Low foetal weights (FW) (inferior tail of predicted High foetal weights (FW) (superior tail
the distribution function). Graphical representation of the distribution function). Graphical
of a sample of 30 real and estimated FW, using a representation of a sample of 38 real and estimated
SVR with polynomial kernel of the 7th grade, a FW, using a SVR with polynomial kernel of the 7th
10% ε-insensitive loss function, regularization grade, a 10% ε-insensitive loss function,
parameter λ= 1000, trained with a separate sub-set regularization parameter λ = 1000, trained with a
of 60 cases. The real foetal weights are ordered separate sub-set of 66 cases. The real foetal
increasingly and represented by dots, and the weights are ordered increasingly and represented
corresponding estimated values are represented by by dots, and the corresponding estimated values are
circles (the lines connecting these circles are for represented by circles (the lines connecting these
visualization purposes only) circles are for visualization purposes only)
Page 3
Acknowledgement
The authors would like to thank to Steve Gunn and
the Image Speech & Intelligent Systems Group,
University of Southampton, United Kingdom, for
letting us experiment the Matlab software
developed for Support Vector Machines for
Classification and Regression.
References
[1] Farmer R.M., Medearis A.L., Hirata G.I., Platt
L.D., 1992, “The Use of a Neural Network for
the Ultrasonographic Estimation of Foetal
Weight in Macrosomic Fetus”, Am J Obstet
Gynecol ,May 1992.
[2] Chauhan S. P. et al., 1998, “Ultrasonographic
estimate of birth weight at 24 to 34 weeks: A
multicenter study”, Am J Obstet Gynecol
October 1998.
[3] Sereno F, Marques de Sá J.P, Matos A,
Bernardes, “The Application of Radial Basis
Functions and Support Vector Machines to the
Foetal Weight Prediction”, in Dagli C H et al.
(eds.), Proceedings of ANNIE ' 2000, Smart
Engineering System Design Conference, St
Louis, 2000.
[4] Vapnik V.N., Statistical Learning Theory,
New York, Springer, 1998.
[5] Cristianini N. & Shawe-Taylor J., An
Introduction to Support Vector Machines: And
Other Kernel-Based Learning Methods,
Cambridge, Cambridge University Press, 2000
[6] Haykin S., Neural Networks - A
Comprehensive Foundation (2d Edition), New
York ,Prentice Hall, 1999.
[7] Cherkassky V, Mulier F., Learning From Data
– Concepts, Theory, and Methods, New York,
John Wiley & Sons, Inc., 1998
[8] Gunn S., Support Vector Machines for
Classification and Regression, Image Speech
& Intelligent Systems Group, University of
Southampton, United Kingdom, 1998.
[9] Evgeniou T, Pontil M, Workshop on support
vector machines, theory and applications,
Center for Biological and Computational
Learning, and Artificial Intelligence
Laboratory, MIT, Cambridge, 2000.
Page 4