TITLE:
Financial Risk Analysis through Malicious URL Detection: Safeguarding Consumer Data in the Digital Economy
AUTHORS:
Soundararajan , Vaibhav Bhaskar, Adith Kadiyala
KEYWORDS:
Machine Learning, Malicious URL Detection, Financial Risk Modeling, Convolutional Neural Network (CNN), ResNet-18, VGG-19, Cybersecurity, Phishing Detection, Identity Theft Prevention, Fraud Detection, Overfitting, Digital Economy, Data Breach Prevention
JOURNAL NAME:
Open Journal of Business and Management,
Vol.13 No.6,
October
28,
2025
ABSTRACT: Using Machine Learning to detect malicious URLs can save millions of people’s data every year and protect them from cyber attackers. This carries major financial consequences, particularly in reducing cases of identity theft, fraudulent banking activity, and exposure of private financial data. By using a Machine Learning-backed model and specific Convolutional Neural Network (CNN) algorithms, we conducted a study that separates user-inputted URLs into distinct categories based on whether they are safe or not. These categories are malware, phishing, defacement, and benign. Categorization of these links enables users to stay safe before clicking on a link, reducing the likelihood of financial loss due to scams or compromised systems. The two specific Machine Learning models used were ResNet 18 and VGG 19. The ResNet 18 model produced a final accuracy of 73 percent, while the VGG 19 model only achieved an accuracy of 34 percent. As shown, ResNet-18 significantly outperformed VGG-19 and delivered strong results despite the data being classified into four distinct categories. Additionally, compared to industry standards for machine learning classification models, our ResNet 18 model performed very well. On the other hand, the VGG 19 model, despite being a highly regarded architecture in the field of computer and data science, did not perform well. Such poor accuracy can be explained by a mathematical phenomenon known as overfitting. However, the main takeaways from this experiment involve the general awareness of online data breaches and prevention strategies, highlighting the model’s potential for application in financial institutions. These models can be applied to real-time detection of suspicious activity, with ResNet-18 showing high accuracy compared to the relatively weaker performance of VGG-19.