Music genre detection using semi-supervised machine learning

Öz

In machine learning, when labeled data is scarce, semi-supervised learning methods are used to improve model performance. In this study, the contribution of self-training, a semi-supervised learning method, is evaluated. In an experimental study with the GTZAN dataset, the effect of self-training on model performance was measured in eight different classifiers. As a result of the experimental studies, it is seen that the self-training method can have a positive effect on model performance depending on certain criteria such as the dataset and the classifier used.

Anahtar Kelimeler

• machine learning
• semi-supervised learning
• self-training

Yarı denetimli makine öğrenmesi yöntemini kullanarak müzik türlerinin tespiti

Öz

Makine öğrenmesinde, etiketli verinin yetersiz olduğu durumlarda, yarı denetimli öğrenme yöntemleri kullanılarak model başarısı artırılmaya çalışılır. Bu çalışmada, bir yarı denetimli öğrenme yöntemi olan kendi kendine öğrenmenin katkısı değerlendirilmiştir. GTZAN veri kümesi ile yapılan deneysel çalışmada, sekiz ayrı sınıflandırıcıda kendi kendine öğrenme yönteminin model başarısına etkisi ölçümlenmiştir. Yapılan deneysel çalışmalar sonucunda, veri kümesi ve kullanılan sınıflandırıcı gibi belirli kriterlerle bağlı olarak kendi kendine öğrenme yönteminin model performansı üzerinde olumlu etkisi olabileceği görülmüştür.

Anahtar Kelimeler

• makine öğrenmesi
• yarı denetimli öğrenme

Kaynakça

1. [1] J. E. van Engelen ve H. H. Hoos, “A survey on semi-supervised learning”, Mach Learn, c. 109, sy 2, ss. 373-440, Şub. 2020, doi: 10.1007/s10994-019-05855-6.
2. [2] M.-R. Amini, V. Feofanov, L. Pauletto, E. Devijver, ve Y. Maximov, “Self-Training: A Survey”. arXiv, 15 Şubat 2023. http://arxiv.org/abs/2202.12040
3. [3] X. Zhu, “Semi-Supervised Learning Literature Survey”, Comput Sci, University of Wisconsin-Madison, c. 2, Tem. 2008.
4. [4] O. Chapelle, B. Schölkopf, ve A. Zien, Ed., Semi-supervised learning. içinde Adaptive computation and machine learning series. Cambridge, Mass. [u.a]: MIT Press, 2010.
5. [5] Ke Chen ve Shihai Wang, “Semi-Supervised Learning via Regularized Boosting Working on Multiple Semi-Supervised Assumptions”, IEEE Trans. Pattern Anal. Mach. Intell., c. 33, sy 1, ss. 129-143, Oca. 2011, doi: 10.1109/TPAMI.2010.92.
6. [6] I. Triguero, S. García, ve F. Herrera, “Self-labeled techniques for semi-supervised learning: taxonomy, software and empirical study”, Knowl Inf Syst, c. 42, sy 2, ss. 245-284, Şub. 2015, doi: 10.1007/s10115-013-0706-y.
7. [7] O. Chapelle ve A. Zien, “Semi-supervised classification by low density separation”, içinde International workshop on artificial intelligence and statistics, PMLR, 2005, ss. 57-64.
8. [8] K. Bennett ve A. Demiriz, “Semi-supervised support vector machines”, Advances in Neural Information processing systems, c. 11, 1998.

9. [9] S. Fralick, “Learning to recognize patterns without a teacher”, IEEE Trans. Inform. Theory, c. 13, sy 1, ss. 57-64, Oca. 1967, doi: 10.1109/TIT.1967.1053952.
10. [10] A. Blum ve T. Mitchell, “Combining labeled and unlabeled data with co-training”, içinde Proceedings of the eleventh annual conference on Computational learning theory, Madison Wisconsin USA: ACM, Tem. 1998, ss. 92-100. doi: 10.1145/279943.279962.
11. [11] Q. Xie, M.-T. Luong, E. Hovy, ve Q. V. Le, “Self-training with Noisy Student improves ImageNet classification”. arXiv, 19 Haziran 2020. http://arxiv.org/abs/1911.04252
12. [12] G. Karamanolakis, S. Mukherjee, G. Zheng, ve A. H. Awadallah, “Self-Training with Weak Supervision”. arXiv, 12 Nisan 2021. http://arxiv.org/abs/2104.05514
13. [13] G. Tzanetakis, “Automatic Musical Genre Classification of Audio Signals.”, Oca. 2001.
14. [14] C. Rosenberg, M. Hebert, ve H. Schneiderman, “Semi-supervised self-training of object detection models”, 2005.
15. [15] N. Kamal, M. Andrew, ve M. Tom, “Semi-Supervised Text Classification Using EM”, içinde Semi-Supervised Learning, O. Chapelle, B. Scholkopf, ve A. Zien, Ed., The MIT Press, 2006, ss. 32-55. doi: 10.7551/mitpress/9780262033589.003.0003.
16. [16] G. Tur, D. Hakkani-Tür, ve R. E. Schapire, “Combining active and semi-supervised learning for spoken language understanding”, Speech Communication, c. 45, sy 2, ss. 171-186, Şub. 2005, doi: 10.1016/j.specom.2004.08.002.
17. [17] D.-H. Lee, “Pseudo-Label : The Simple and Efficient Semi-Supervised Learning Method for Deep Neural Networks”, ICML 2013 Workshop : Challenges in Representation Learning (WREPL), Tem. 2013.
18. [18] Y. Zou, Z. Yu, B. V. K. V. Kumar, ve J. Wang, “Domain Adaptation for Semantic Segmentation via Class-Balanced Self-Training”, 2018, doi: 10.48550/ARXIV.1810.07911.
19. [19] P. Cascante-Bonilla, F. Tan, Y. Qi, ve V. Ordonez, “Curriculum Labeling: Revisiting Pseudo-Labeling for Semi-Supervised Learning”. arXiv, 10 Aralık 2020. http://arxiv.org/abs/2001.06001
20. [20] K. Sohn vd., “FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence”. arXiv, 25 Kasım 2020. http://arxiv.org/abs/2001.07685
21. [21] P. Yilmaz, Ş. Akçakaya, Ş. D. Özkaya, ve A. Çeti̇N, “Machine Learning Based Music Genre Classification and Recommendation System”, ECJSE, Ara. 2022, doi: 10.31202/ecjse.1209025.
22. [22] S. Sigtia ve S. Dixon, “Improved music feature learning with deep neural networks”, içinde 2014 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Florence, Italy: IEEE, May. 2014, ss. 6959-6963. doi: 10.1109/ICASSP.2014.6854949.
23. [23] V. Kiranoglu, G. Tüysüzoğlu, ve E. Öztürk Kiyak, “Prediction of Crime Occurrence in case of Scarcity of Labeled Data”, Deu Muhendislik Fakultesi Fen ve Muhendislik, c. 23, sy 68, ss. 677-687, May. 2021, doi: 10.21205/deufmd.2021236828.
24. [24] I. Triguero, J. A. Sáez, J. Luengo, S. García, ve F. Herrera, “On the characterization of noise filters for self-training semi-supervised in nearest neighbor classification”, Neurocomputing, c. 132, ss. 30-41, May. 2014, doi: 10.1016/j.neucom.2013.05.055.
25. [25] Y. Wang vd., “USB: A Unified Semi-supervised Learning Benchmark for Classification”. arXiv, 13 Ekim 2022. http://arxiv.org/abs/2208.07204
26. [26] B. Zhang vd., “FlexMatch: Boosting Semi-Supervised Learning with Curriculum Pseudo Labeling”. arXiv, 28 Ocak 2022. http://arxiv.org/abs/2110.08263
27. [27] B. Zoph vd., “Rethinking Pre-training and Self-training”. arXiv, 15 Kasım 2020. http://arxiv.org/abs/2006.06882
28. [28] M. Li ve Z.-H. Zhou, “SETRED: Self-training with Editing”, içinde Advances in Knowledge Discovery and Data Mining, c. 3518, T. B. Ho, D. Cheung, ve H. Liu, Ed., içinde Lecture Notes in Computer Science, vol. 3518. , Berlin, Heidelberg: Springer Berlin Heidelberg, 2005, ss. 611-621. doi: 10.1007/11430919_71.
29. [29] Y. Zou, Z. Yu, X. Liu, B. V. K. V. Kumar, ve J. Wang, “Confidence Regularized Self-Training”. arXiv, 15 Temmuz 2020. http://arxiv.org/abs/1908.09822
30. [30] A. Krizhevsky, G. Hinton, ve others, “Learning multiple layers of features from tiny images”, 2009.
31. [31] H. Schmutz, O. Humbert, ve P.-A. Mattei, “Don’t fear the unlabelled: safe semi-supervised learning via simple debiasing”. arXiv, 03 Mart 2023. http://arxiv.org/abs/2203.07512
32. [32] A. Olteanu, “GTZAN Dataset-Music Genre Classification. Kaggle. com (2019)”. https://www.kaggle.com/datasets/andradaolteanu/gtzan-dataset-music-genre-classification
33. [33] T. Hastie, J. Friedman, ve R. Tibshirani, The Elements of Statistical Learning. içinde Springer Series in Statistics. New York, NY: Springer New York, 2001. doi: 10.1007/978-0-387-21606-5.
34. [34] T. Cover ve P. Hart, “Nearest neighbor pattern classification”, IEEE Trans. Inform. Theory, c. 13, sy 1, ss. 21-27, Oca. 1967, doi: 10.1109/TIT.1967.1053964.
35. [35] B. Charbuty ve A. Abdulazeez, “Classification Based on Decision Tree Algorithm for Machine Learning”, JASTT, c. 2, sy 01, ss. 20-28, Mar. 2021, doi: 10.38094/jastt20165.
36. [36] S. J. Rigatti, “Random Forest”, Journal of Insurance Medicine, c. 47, sy 1, ss. 31-39, Oca. 2017, doi: 10.17849/insm-47-01-31-39.1.
37. [37] M. Belgiu ve L. Drăguţ, “Random forest in remote sensing: A review of applications and future directions”, ISPRS Journal of Photogrammetry and Remote Sensing, c. 114, ss. 24-31, Nis. 2016, doi: 10.1016/j.isprsjprs.2016.01.011.
38. [38] T. Chen ve C. Guestrin, “XGBoost: A Scalable Tree Boosting System”, Mar. 2016.
39. [39] D. Mustafa Abdullah ve A. Mohsin Abdulazeez, “Machine Learning Applications based on SVM Classification A Review”, QAJ, c. 1, sy 2, ss. 81-90, Nis. 2021, doi: 10.48161/qaj.v1n2a50.
40. [40] M. Maalouf, “Logistic regression in data analysis: an overview”, IJDATS, c. 3, sy 3, s. 281, 2011, doi: 10.1504/IJDATS.2011.041335.
41. [41] J. Zou, Y. Han, ve S.-S. So, “Overview of Artificial Neural Networks”, içinde Artificial Neural Networks, c. 458, D. J. Livingstone, Ed., içinde Methods in Molecular BiologyTM, vol. 458. , Totowa, NJ: Humana Press, 2008, ss. 14-22. doi: 10.1007/978-1-60327-101-1_2.
42. [42] Nart Sooksil ve Vacharapoom Benjaoran, “Non-linear modelling of construction workers’ behaviorsfor accident prediction”, Songklanakarin Journal of Science and Technology (SJST), c. 43, s. 596602, 2021, doi: 10.14456/SJST-PSU.2021.80.