A COMPREHENSIVE FRAMEWORK FOR DETECTING BOT-DRIVEN FRAUDULENT TRANSACTIONS IN FINTECH AND ECOMMERCE COMPANIES
Abstract - As bot attacks increasingly compromise the cybersecurity of fintech and eCommerce companies, there's an urgent need to develop more effective defense strategies. These attacks lead to significant financial losses through credit write-offs from fraudulent transactions and risk compromising sensitive user data. This paper proposes a comprehensive framework to counter such threats by detecting bot-driven fraudulent transactions. The framework incorporates an automated bot detection system leveraging machine learning (ML), which enhances detection speed and response time, providing cleaner login data for informed decision-making. This paper further discusses the limitations of existing statistical anomaly detection techniques, emphasizing the need for more complex solutions. In response, we propose the integration of causal inference-based methods, utilizing ML models for generating counterfactuals and segment-specific bot activity flagging. The framework also layers human knowledge onto the ML approach, prioritizing segments driving the majority of transactions, and considering the triggering of risk models as ground truth for bot activity detection. Finally, we address the scalability and generalizability of the framework across different products and metrics, underscoring its potential for wide adoption in enhancing cybersecurity measures in fintech and eCommerce companies.
Keywords - Automated Bot Attack Detection, Machine Learning Framework, Applied Causal Inference, Fraud Transactions, Cybersecurity