![](bg4.png)
Table 2: Genetically optimised SVM parameter for human brain dataset using GA-based approach and Grid algorithm.
Test period
GA-based approach Grid algorithm
C
Accuracy Training days C
Accuracy Training days
TP1 8 0.725 86.9565 % 10 0.5 0.0625 58.33 % 0.2
TP2 2383 0.0008 98.5682 % 8 7 2 97.19 % 0.8
TP3 19277 0.0087 99.8897 % 11 3 0.9 91.33 % 0.5
TP4 133455 0.2502 100 % 11 9 1.1245 95.33 % 0.6
TP5 131847 0.0009 100 % 11 8 2.019 94.71 % 0.9
The optimised values of SVM parameters for human
brain dataset using GA-based approach and Grid
algorithm corresponding to each TP are given in
“Table 2”. It can be observed that the optimum
values of these parameters vary significantly over a
wide range reflecting the superiority of GA to Grid
algorithm. In the GA, pairs of (C, γ) are tried and the
one with the best accuracy is chosen. To obtain the
best optimized pair of (C, γ), the process lasts
between 8 and 11 days, but the best accuracy is
achieved with a longer period. The parameters C
whose values exceed 2000 achieve high accuracy
surpassing 98% to 100%. In the Grid algorithm the
accuracy rate is low despite the short period of
training. Comparison of the obtained results of GA
with those of Grid algorithm demonstrates that GA-
SVM approach has a better classification accuracy
than the Grid algorithm tested.
5 CONCLUSIONS
This study presents an evolutionary computing
optimization approach, capable of searching for the
optimal parameter values for SVM by using a subset
of selected features. Compared with the statistical
approach, the proposed GA-based approach has
higher accuracy with fewer selected features. It
outperforms the statistical approach in terms of
computational efficiency. Moreover, the proposed
GA-based approach has proved to be effective in
optimizing parameters for the SVM. Results of this
study are obtained with an RBF kernel function.
However, other kernel parameters can also be
optimized using the same approach. This is of
particular significance to medical decision in the
medical diagnostic field.
For future work, we intend to add coefficient of
ponderation for each of the five selected features.
We would also to extend our approach to real-world
problems and other public datasets such as heart
disease and breast cancer.
REFERENCES
Kharrat A., Gasmi K., Ben Messaoud M., Benamrane N.,
Abid M., 2010a. Automated Classification of
Magnetic Resonance Brain Images Using Wavelet
Genetic Algorithm and Support Vector Machine. Proc.
9th IEEE Int. Conf. on Cognitive Informatics
(ICCI’10), doi: 10.1109/COGINF.2010.5599712. 369-
-374.
Kharrat A., Ben Messaoud M., Benamrane N., Abid M.,
2010b. Genetic Algorithm for Feature Selection of
MR Brain Images Using Wavelet Co-occurrence.
Proc. IEEE Int. Conf. on Signal and Information
Processing (ICSIP 2010), 1--5.
Boser B. E., Guyon I. M., Vapnik V. N., 1992. A training
algorithm for optimal margin classifiers. Proceedings
of the 5th Annual ACM Workshop on Computational
Learning Theory, 144--152.
Kharrat A., Gasmi K., Ben Messaoud M., Benamrane N.,
Abid M., 2010c. A Hybrid approach for automatic
classification of brain MRI using genetic algorithm
and support vector machine. Leonardo Journal of
Sciences (LJS), 9, 71--82.
Raymer, M. L., Punch, W. F., Goodman, E. D., Kuhn, L.
A., Jain, A. K., 2000. Dimensionality reduction using
genetic algorithms. IEEE Transactions on
Evolutionary Computation, 4, 164--171.
Yang, J., Honavar, V., 1998. Feature subset selection
using a genetic algorithm. IEEE Intelligent Systems,
13, 44--49.
Salcedo-Sanz, S., Prado-Cumplido, M., Pérez-Cruz, F.,
Bousono-Calzon, C., 2002. Feature selection via
genetic optimization. Proceedings of the ICANN
international conference on artificial neural networks,
547--552.
Frohlich, H., Chapelle, O., 2003. Feature selection for
support vector machines by means of genetic
algorithms. Proceedings of the 15th IEEE
international conference on tools with artificial
intelligence, Sacramento, 142--148.
Keith, A. J., Alex, J. B., 1999. The whole brain atlas.
Harvard Medical School, Boston. Available from:
http://med.harvard.edu/AANLIB/.
VISAPP 2011 - International Conference on Computer Vision Theory and Applications
204