پیش بینی احتمال وقوع ریسک عملیاتی در صنعت بانکداری با استفاده از الگوریتم های یادگیری ماشین (مقاله علمی وزارت علوم)
درجه علمی: نشریه علمی (وزارت علوم)
آرشیو
چکیده
این پژوهش با هدف پیش بینی احتمال وقوع ریسک عملیاتی در صنعت بانکداری با استفاده از الگوریتم های یادگیری ماشین انجام شده است. پژوهش حاضر با تحلیل داده های ریسک عملیاتی و ارزیابی عملکرد الگوریتم های یادگیری ماشین به منظور ارائه الگوریتم هایی مؤثر برای پیش بینی دقیق تر احتمال وقوع ریسک عملیاتی و مدیریت بهتر آن در صنعت بانکداری صورت گرفته است. در این پژوهش، داده های مرتبط با ریسک عملیاتی از سال 1395 تا 1402 جمع آوری و پیش پردازش شد و سپس با استفاده از مدل های یادگیری ماشین مانند RF، DT، SVM، LR، NB و KNN پیش بینی انجام شد. عملکرد مدل ها با معیارهایی همچون دقت، صحت، بازخوانی، F1-score و AUC ارزیابی شد تا بهترین مدل برای پیش بینی احتمال وقوع ریسک انتخاب شود. نتایج نشان می دهند که الگوریتم های RF و SVM در پیش بینی احتمال وقوع ریسک عملیاتی در تمامی حالت ها عملکرد بسیار خوبی دارند؛ به علاوه که الگوریتم های یادگیری ماشین توانایی بالایی در پیش بینی وقوع ریسک عملیاتی دارند و می توانند ابزار مؤثری برای تصمیم گیری های مدیریتی در صنعت بانکداری فراهم کنند.Predicting the Likelihood of Operational Risk Occurrence in the Banking Industry Using Machine Learning Algorithms
This study investigates and predicts the likelihood of operational risk occurrence in the banking industry using machine learning algorithms. The primary objective is to analyze operational risk data and evaluate the performance of various machine learning models to develop effective tools for enhancing risk management and minimizing financial losses in banks and financial institutions. Operational risk data were collected, pre-processed, and then used for predictions with machine learning models, including Random Forest (RF), Decision Tree (DT), Support Vector Machine (SVM), Logistic Regression (LR), Naïve Bayes (NB), and k-Nearest Neighbors (KNN). Model performance was assessed using evaluation metrics such as accuracy, precision, recall, F1-score, and the Area Under the Curve (AUC) to determine the most effective model for risk prediction. The findings indicate that the RF and SVM algorithms outperform other models in predicting operational risk across all scenarios. Furthermore, the results demonstrate the strong predictive capability of machine learning algorithms in assessing operational risk, highlighting their potential as valuable decision-making tools for risk management in the banking sector. Keywords: Risk Prediction, Operational Risk, Risk Management, Machine Learning Introduction Operational risk is defined as the risk arising from external factors or failures in internal controls or information systems, which may lead to both anticipated and unexpected losses (Crouchy et al., 1998). Lopez (2002) characterizes it as any unquantifiable risk that a bank may encounter. According to the Basel II Agreement, operational risk refers to the probability of loss resulting from deficiencies, breakdowns, or inefficiencies in human resources, processes, technologies, infrastructure, or internal and external events (Pena et al., 2018). To estimate the capital required to cover operational risk, the Basel framework introduces three approaches: the Basic Indicator Approach (BIA), the Standardized Approach (SA), and the Advanced Measurement Approach (AMA) (Mora Valencia, 2010; Mora Valencia et al., 2017). The BIA and SA estimate capital requirements based on annual gross income, with the key distinction being that the SA categorizes a bank’s activities into eight business lines. Under the BIA, an alpha coefficient (α) of 15% is applied, whereas in the SA, each business line has a specific beta coefficient (β) ranging between 12% and 18%. The AMA employs both quantitative and qualitative methods for operational risk modeling, leveraging databases to collect statistical data and utilizing the loss distribution approach (LDA) to model frequency and severity distributions. Capital coverage is then determined based on the cumulative distribution of these variables. Since the LDA is data-driven, the Basel framework (BCBS, 2004) emphasizes the necessity of a robust database for collecting operational risk data. Four key databases are required: internal loss event data, external loss event data, scenario-based analysis data, and a database of business environment and internal control factors. Compared to other banking risks, such as credit and market risks, measuring, monitoring, and managing operational risk is considerably more complex. This risk has gained increasing attention in recent years, as large operational losses have led to the liquidation of financial institutions (Abdymomunov et al., 2020; Afonso et al., 2019). Crisanto and Perino (2017) identify cyber threats and cyber fraud as critical factors influencing operational risk capital estimation. These risks have intensified with the growth of electronic banking services and include illegal access, system disruptions, and the misuse or theft of digital assets for financial gain (BCBS, 2016; Drew & Farrell, 2018). To quantify potential losses in electronic banking transactions, Bouveret (2018) proposed a Bayesian Network (BN) model to estimate operational risk capital requirements in financial institutions. Machine learning has emerged as one of the most promising yet challenging approaches in modern finance (Tsai & Wu, 2008). These methods have transformed the financial industry, with deep learning (DL) being extensively studied and applied due to its adaptability and predictive capabilities (Ivanov, 2019). Pena et al. (2021) employed a fuzzy convolutional deep learning model to estimate the maximum operational risk value at a 99.9% confidence level. Similarly, Zhou et al. (2020) utilized semi-supervised machine learning algorithms to classify operational risks based on financial news, analyzing 5,843 documents from financial articles and newspapers in the Asia-Pacific region between February and March 2019. Their model demonstrated the capability to predict various types of risks in the banking industry. In another study, Akbari and Yazdanian (2023) applied machine learning algorithms to determine optimal thresholds for operational loss severity data, classifying the data and estimating the capital required to cover operational risk by integrating severity and frequency distribution functions with Monte Carlo simulation. Method and Data In this study, operational risk data were collected, pre-processed, and then used for predictions with machine learning models, including RF, DT, SVM, LR, NB, and KNN. The models' performance was assessed using evaluation metrics such as accuracy, precision, recall, F1-score, and AUC to identify the most effective model for predicting the likelihood of risk occurrence. Findings The results indicate that the RF and SVM algorithms exhibit strong performance in predicting operational risk across all scenarios. Specifically, the RF algorithm achieved an accuracy of 0.9690, while the SVM algorithm attained an accuracy of 0.9587 in State 1, making them the most effective models in this setting. Both algorithms demonstrated comparable performance across other modes. Conclusion and Discussion This study analyzes and predicts operational risk occurrence in the banking industry using machine learning algorithms. The findings indicate that various algorithms, particularly RF and SVM, demonstrate strong predictive performance. These results have the potential to transform operational risk management in banks, leading to significant reductions in associated costs and losses. A key insight from this study is that leveraging large and diverse datasets can substantially enhance prediction accuracy. Machine learning models can process complex datasets, identify hidden patterns, and facilitate early risk detection, enabling banks to implement preventive measures before risks materialize. Moreover, integrating machine learning into risk management enhances decision-making by providing precise, data-driven predictions, allowing for more effective strategies and efficient resource allocation. Future research could incorporate additional data, such as historical records, economic indicators, and internal process information, to further improve prediction accuracy. With advancements in technology, more sophisticated techniques—such as reinforcement learning methods (e.g., DQN, Q-Learning, DDPG, and Meta-Learning)—could enhance the accuracy and efficiency of operational risk prediction models.
تبلیغات
