ÇOK KATMANLI PERSEPTRON SİNİR AĞLARI İLE DİYABET HASTALIĞININ TEŞHİSİ

Öz

Yapay sinir ağları (YSA), farklı disiplinlerdeki karmaşık problemlerin çözümlenmesinde kabul gören ve uygulamalarda sıklıkla yer alan modelleme araçları haline gelmiştir. Farklı YSA yapıları, tıp alanında karar destek sistemlerinin gelişmesinde kullanılmakta olan önemli modellerdendir. Bu çalışmada, dört farklı algoritma ile eğitilen çok katmanlı perseptron sinir ağları diyabet hastalığının teşhisinde kullanılmış ve en başarılı algoritma belirlenmiştir. Geri yayılım, delta-bar-delta, genişletilmiş delta-bar-delta ve hızlı yayılım, çalışılmış olan dört algoritmadır. Çok katmanlı perseptron sinir ağlarının eğitimi, geçerliliği ve testi veri tabanında yer alan farklı kişilere ait kayıtlar ile yapılmıştır. Performans belirleyiciler ve istatistiksel ölçümler ile çok katmanlı perseptron sinir ağları değerlendirilmiş ve sonuçlar diyabet hastalığının teşhisinde hızlı yayılım algoritmasının en
başarılı çok katmanlı perseptron eğitim algoritması olduğunu göstermiştir.

Anahtar Kelimeler

• Çok katmanlı perseptron sinir ağı, eğitim algoritmaları, diyabet hastalığının teşhisi.

DIABETES DIAGNOSIS BY MULTILAYER PERCEPTRON NEURAL NETWORKS

Öz

Artificial neural networks (ANNs) have become modeling tools that have found extensive acceptance and they have frequently used in applications in many disciplines for solving complex problems. Different ANN structures are valuable models, which are used in the medical field for the development of decision support systems. In this study, four multilayer perceptron neural networks (MLPNNs) trained with different algorithms were used for diabetes prediction and the most efficient training algorithm was determined. Backpropagation, delta-bar-delta, extended delta-bar-delta and quick propagation were the studied four training algorithms. The MLPNNs were trained, cross validated and tested with subject records from the database. Performance indicators and statistical measures were used for evaluating the MLPNNs and the results demonstrated that the quick propagation algorithm was the most efficient multilayer perceptron training algorithm for diabetes prediction. 

Anahtar Kelimeler

• Multilayer perceptron neural network,training algorithms,diabetes diagnosis

Kaynakça

1. Tafeit, E., Reibnegger, G., “Artificial Neural
2. Networks in Laboratory Medicine and Medical
3. Outcome Prediction”, Clinical Chemistry and
4. Laboratory Medicine, Vol 37, No 9, 845-853, 1999.
5. Güler, İ., Übeyli, E.D., “Detection of Ophthalmic
6. Artery Stenosis by Least-Mean Squares
7. Backpropagation Neural Network”, Computers in
8. Biology and Medicine, Vol 33, No 4, 333-343, 2003.

9. Übeyli, E.D., Güler, İ., “Neural Network Analysis
10. of Internal Carotid Arterial Doppler Signals:
11. Predictions of Stenosis and Occlusion”, Expert
12. Systems with Applications, Vol 25, No 1, 1-13, 2003.
13. Güler, İ, Übeyli, E.D., “ECG Beat Classifier
14. Designed by Combined Neural Network Model”,
15. Pattern Recognition, Vol 38, No 2, 199-208, 2005.
16. Haykin, S., Neural Networks: A Comprehensive
17. Foundation, New York, Macmillan College
18. Publishing Company Inc, 1994.
19. Chaudhuri, B.B., Bhattacharya, U., “Efficient
20. Training and Improved Performance of
21. Multilayer Perceptron in Pattern Classification”,
22. Neurocomputing, Vol 34, 11-27, 2000.
23. Rumelhart, D.E., Hinton, G.E., Williams, R.J.,
24. “Learning Representations by Back-Propagating
25. Errors”, Nature, Vol 323, 533-536, 1986.
26. Jacobs, R.A., “Increased Rate of Convergence
27. through Learning Rate Adaptation”, Neural
28. Networks, Vol 1, 295-307, 1988.
29. Minai, A.A., Williams, R.D., “Back-Propagation
30. Heuristics: A Study of the Extended Delta-Bar-
31. Delta Algorithm”, Proceedings of International
32. Joint Conference on Neural Networks, Vol 1,
33. -600, San Diego, California, 17-21 June, 1990.
34. Fahlman, S.E., “An Empirical Study of Learning
35. Speed in Backpropagation Networks”, Computer
36. Science Technical Report, CMU-CS-88-162,
37. Carnegie Mellon University, Pittsburgh, 1988.
38. Chan, L-W., “Efficacy of Different Learning
39. Algorithms of the Back Propagation Network”,
40. IEEE Region 10 Conference on Computer and
41. Communication Systems, Vol 1, 23-27, 24-27
42. September, 1990.
43. Sidani, A., Sidani, T., “A Comprehensive Study
44. of the Back Propagation Algorithm and
45. Modifications”, IEEE Conference Record, 80-
46. , 29-31 March, 1994.
47. Hannan, J.M., Bishop, J.M., “A Comparison of
48. Fast Training Algorithms Over Two Real
49. Problems”, IEE Fifth International Conference
50. on Artificial Neural Networks, Conference
51. Publication No. 440, 1-6, 7-9 July, 1997.
52. Download Diabetes Data; 05.05.2005.
53. http://www.cormactech.com/neunet/download.html,
54. Besser, G.M., Bodansky, H.J., Cudworth A.G.,
55. Clinical Diabetes an Illustrated Text, Gower
56. Medical Publishing, London, 1988.
57. Kennedy, L., Lyons, T.J., “Non-Enzymatic
58. Glycosylation”, British Medical Bulletin, Vol
59. , No 1, 174-190, 1989.
60. Bray, G.A., Contemporary Diagnosis and
61. Management of Obesity, Newtown, Pennsylvania,
62. Handsbook in Health Care Company, 2003.
63. Hyöty, H., Hiltunen, M., Lönnrot, M., “Enterovirus
64. Infections and Insulin Dependent Diabetes
65. Mellitus-Evidence for Causality”, Clinical and
66. Diagnostic Virology, Vol 9, No 2-3, 77-84, 1998.
67. Pi-Sunyer, F.X., “A Review of Long-Term
68. Studies Evaluating the Efficacy of Weight Loss in
69. Ameliorating Disorders Associated with Obesity”,
70. Clinical Therapeutics, Vol 18, No 6, 1006-1035,
71. -
72. American Diabetes Association (ADA), 05.05.2005
73. http://www.diabetes.org/home.jsp