DENETİMLİ MAKİNE ÖĞRENMESİ TEKNİKLERİNİ KULLANARAK FİNANSAL BİLGİ MANİPÜLASYONUNUN TESPİTİ: SVM, PNN, KNN, DT

Bu çalışma kapsamında, finansal bilgi manipülasyonunu tahmin etmek için geleneksel tahmin algoritmaları ve denetimli makine öğrenmesi yöntemleri kullanılmaktadır. Geleneksel tahmin algoritması olarak logit kullanılırken, denetimli makine öğrenmesi yöntemlerinden destek vektör makinesi (SVM), olasılıksal sinir ağı (PNN), k-en yakın komşu (KNN) ve karar ağacı (DT) algoritmaları kullanılmıştır. Önceki çalışmalara göre, destek vektör makinesi ve olasılıksal sinir ağı algoritmaları geleneksel tahmin algoritmalarından daha yüksek performans göstermektedir. 2009-2018 yılları arasında Sermaye Piyasası Kurulu'nun haftalık bültenlerini gözden geçirerek toplanan verilere tüm algoritmalar ayrı ayrı uygulanmıştır. Hangi algoritmanın finansal bilgi manipülasyonunu tespitinde daha başarılı olduğuna karar vermek amacıyla karşılaştırmalı analiz yapılmıştır. Karşılaştırmalı analizde, algoritmaların duyarlılık ve özgünlük istatistiklerinin performansına bakılmıştır. Elde edilen sonuçlar, KNN ve SVM’nin diğer algoritmalardan daha iyi performansa sahip olduğunu ve kullanılan tüm algoritmaların önceki literatürün sonuçlarına kıyasla yüksek performansa sahip olduğunu göstermektedir.

Anahtar Kelimeler:

Finansal Bilgi Manipülasyonu, Denetimli Makine Öğrenmesi, SVM, KNN, Beneish

DETECTING FINANCIAL INFORMATION MANIPULATION BY USING SUPERVISED MACHINE LEARNING TECHNICS: SVM, PNN, KNN, DT

Within the scope of this paper, traditional estimation algorithms and supervised machine learning methods are used to estimate the manipulation of financial information. Traditional estimation algorithms, such as logit, and supervised machine learning methods, which are support vector machine (SVM), probabilistic neural network (PNN), k-nearest neighbor (KNN) and decision tree (DT) algorithms, are utilized. According to previous studies, support vector machine and probabilistic neural network algorithms perform higher than traditional estimation algorithms. Comparative analysis is made to decide better algorithm for classification by applying all algorithms separately to the data that is collected by skimming weekly bulletins of Capital Market Board of Turkey between 2009 and 2018. Thus, it is determined which algorithms perform better in financial information manipulation by looking at performance of classification accuracy, sensitivity and specificity statistics. The obtained results show that KNN and SVM have better performance than the other algorithms and all utilized algorithms have high performance compared to the previous literature’s results.

Keywords:

Financial Information Manipulation, Supervised Machine Learning, SVM, KNN, Beneish,

PDF

___

Aktaş, R., Alp, A. and Doğanay, M. M. (2007). Towards predicting financial information manipulation. The ICFAI Journal of Applied Finance, 13(7), 39–52.
Arens, A. A. and Loebbecke, J. K. (2000). Auditing: An integrated approach (Internatial ed.). New Jersey: Prentice Hall International, Inc.
Beneish, Messod D. (1999). The Detection of Earnings Manipulation. Financial Analysts Journal, 55(5), 24–36.
Cortes, Corinna., Vapnik, Vladimir N. (1995). Support-Vector Network. Machine Learning, 20, 273-297.
DataCamp. (2018). Decision Tree Classification in Python. Retrieved from www.datacamp.com/community/tutorials/decision-tree-classification-python.
Doğanay, M., Aktaş, R., Alp, A. and Öğüt H. (2009). Prediction of Financial Information Manipulation by Using Support Vector Machine and Probabilistic Neural Network. Expert Systems with Applications, 36, 5419–5423.
Fanning, K., M. and Cogger, K., O. (1998). Neural Network Detection of Management Fraud Using Published Financial Data, International Journal of Intelligent Systems in Accounting, Finance & Management, 7(1), 21-41.
Hsu, C.-W., Chang, C.-C. and Lin, C.-J. (2004). A Practical Guide to Support Vector Classification, Technical Report. Department of Computer Science and Information Engineering, National Taiwan University.
Küçüksözen, C. (2004). Financial Information Manipulation: Causes, Methods, Objectives, Techniques, Results And An Empirical Study On IMKB Companies. (Unpublished Doctoral Dissertation). Ankara University Social Science Institute, Management, Ankara.
Liou, F. M. (2008). Fraudulent Financial Reporting Detection And Business Failure Prediction Models: A Comparison. Managerial Auditing Journal, 23(7), 650-662.
Machine Learning Mastery. (2019). A Gentle Introduction to k-Fold Cross-Validation. Retrieved from machinelearningmastery.com/k-fold-cross-validation/.
Moise, I., Pournaras, E. and Helbing, D. (2015). K-Nearest Neighbour Classifier.
NeuPy (2019). Probabilistic Neural Network (PNN). Retrieved from neupy.com/apidocs/neupy.algorithms.rbfn.pnn.html.
OECD (2004). OECD Principles of Corporate Governance.
Rokach, L. and Maimon, O. (2005). Top-Down Induction of Decision Trees Classifiers—A Survey. IEEE Trans. Syst., Man, Cybern. C, Appl. Rev, 35(4), 476–487.
Rutkowski, L. (2004). Adaptive Probabilistic Neural Networks for Pattern Classification In Time-Varying Environment. IEEE Transactions on Neural Networks, 15, 811–827.
Scikit (2018). Sklearn.feature_selection.RFE. Retrieved from scikit-learn.org/stable/modules/generated/sklearn.feature_selection.RFE.html.
Scikit (2019). Sklearn.neighbors.KNeighborsClassifier. Retrieved from scikit-learn.org/stable/modules/generated/sklearn.neighbors.KNeighborsClassifier.html#sklearn.neighbors.KNeighborsClassifier.kneighbors.
Spathis, C. T. (2002). Detecting False Financial Statements Using Published Data: Some Evidence From Greece. Managerial Auditing Journal, 17(4), 179-191.
Tibshirani, R. (2019). K-Fold Cross-Validation. Cross-Validation and Bootstrap. Retrieved from statweb.stanford.edu/~tibs/sta306bfiles/cvwrong.pdf.
Vincent, C., Kevin, C. (2002). An introduction to Probabilistic Neural Networks.