Наука и технология в современном миреМирзаева Ширин Нодировна, Тахиров Чингиз Аъзамович +22026
Comparing Neural Networks and Traditional Algorithms in Fraud Detection
PAGE NO. - · Comparing Neural · Algorithms in Fraud · Dip Bharatbhai Patel
The American Journal of Applied sciences · 07 · 2025
Аннотация
Fraud detection has become an essential component of financial security systems. Traditional OPEN ACCESS algorithms have long served as the backbone of these SUBMITED 18 June 2025 systems. The rise of neural networks is revolutionizing ACCEPTED 24 June 2025 PUBLISHED 27 July 2025 the process as it offers new approaches to identifying VOLUME Vol.07 Issue 07 2025 complex fraud patterns. The paper presents a comparative analysis of neural networks and traditional CITATION algorithms. These include decision trees, rule-based Dip Bharatbhai Patel. (2025). Comparing Neural Networks and Traditional Algorithms in Fraud Detection. The American Journal of systems, and logistic regression in fraud detection. The Applied Sciences, 7(07), 128–132. comparison is based on scalability, accuracy, https://doi.org/10.37547/tajas/Volume07Issue07-13 interpretability, computational efficiency, and adaptability. The findings reveal that neural networks COPYRIGHT outperform traditional methods in subtle, non-linear © 2025 Original content from this work may be used under the terms of the creative commons attributes 4.0 License. fraud patterns but suffer from interpretability and data requirements. A hybrid detection framework that combines neural intelligence with rule-based logic is proposed for real-time, robust fraud management. For instance, a neural ensemble model achieved over 97% accuracy while traditional systems achieved 89-91%. The paper highlights that the hybrid approach offers optimal results in real-world scenarios. Keywords: Fraud detection, neural networks, machine learning, traditional algorithms, anomaly detection. I. INTRODUCTION Fraud poses a significant threat to global financial systems. It is costing many companies billions annually. As digital transactions increase, the complexity of fraudulent behaviors also increases. Therefore, fraud detection systems must evolve to detect and identify fraudsters' ever-changing tactics. Historically, traditional algorithms have served as the key detection mechanism. However, neural networks and deep learning have The American Journal of Applied Sciences 128 https://www.theamericanjournals.com/index.php/tajas emerged as the most effective tools in predictive computationally expensive on large datasets. We have analytics and anomaly detection. Traditional machine Rule-based systems that heavily rely on domain learning techniques like decision trees, support vector expertise. Rule-based systems are critical to providing machines, and logistic regression are widely used due to transparent decisions. However, these models are their transparency, and they are easy to deploy. Neural brittle and demand frequent updates to match the ever- networks have revolutionized the field of pattern changing fraud patterns. recognition and anomaly detection. It is capable of Okur et al. (2021) believe the Support Vector Machine learning complex temporal, behavioral, and spatial (SVM) model is effective in high-dimensional spaces. patterns. This strategy is effective in adapting to fraud They are used in linear and non-linear classification. tactics that escape traditional logic. However, they have limitations, such as handling large- Despite these strengths, neural models are often scale datasets. These authors discovered that rule-based criticized for their “black box” nature and computational and statistical models like decision SVM and decision complexity. The paper compares traditional algorithms trees are still being used extensively. Their simplicity and neural networks to determine their strengths and allows for rapid deployment and transparency. They, weaknesses in fraud detection. It will explore how each however, limit adaptability to changing fraud trends. method works, analyze performance metrics, and assess B. Neural Networks for Fraud Detection their practicality in real-world applications. Neural networks, intense learning models such as RNNs II. Literature Review and CNNs, improve fraud detection accuracy. They are Fraud detection involves identifying suspicious or capable of automatically learning hierarchical data unauthorized transactions within massive datasets. patterns. Osegi and Jumbo (2021) propose a simulated Techniques used can be categorized into classifications, annealing-trained neural model outperforming which are supervised. We have unsupervised, which is traditional classifiers. anomaly detection. Traditional methods like decision Artificial Neural Networks (ANNs) trees, support vector machines, and logistic regression These networks mimic brain-like structures, and they rely on predefined rules and feature engineering. Neural can handle complex non-linear relationships. These networks, such as deep learning models, leverage models are effective when they are well-trained on large hierarchical layers to learn complex data datasets. These datasets must be well-labeled to make representations. it easy to operate. Many issues are encountered by fraud detection Co
Доказательный фрагмент
«Fraud detection has become an essential component of financial security systems.»
Образец извлечения
Демо-превьюFraud detection has become an essential component of financial security systems. Полный текст публикации CiteX не хранит — доступ к первоисточнику открывается по DOI согласно лицензии издателя и выбранному плану.
- Цитирований
- 0
- Тип
- Журнал
- Страна
- Узбекистан
- Ключевые слова
- Fraud detection, neural networks, machine, traditional algorithms, machine learning
Смежные источники
Наука и технология в современном миреHamdamov Islom Akramovich2026
Медицина и инновацииМ Сабиров, О Скосырева +12021