Research Projects

This dissertation advances robust anomaly detection through cutting-edge Explainable AI (XAI) techniques, addressing critical challenges such as adversarial attacks, bias mitigation, and concept drift across diverse datasets from domains like cyber security, healthcare, and e-commerce. By leveraging XAI to diagnose vulnerabilities, ensure fairness, and adapt to dynamic data distributions, this research enhances model reliability, interpretability, and resilience against evolving threats. The proposed approaches demonstrate scalability and generalizability, paving the way for secure, trustworthy AI systems that deliver actionable insights and redefine standards in anomaly detection. 

The rise of NFTs, Web3, and decentralized technologies has transformed financial ecosystems, with influencers on platforms like Twitter and Instagram playing a central role in promoting blockchain-based projects and shaping investment behaviors. During the 2020–2022 Web3 craze, numerous projects collapsed shortly after being endorsed by influencers, raising critical questions about accountability and trust in digital finance. This paper explores the extent to which influencers are held accountable for flawed financial advice, using Natural Language Processing (NLP) to analyze social media posts, audience reactions, and engagement metrics following significant price crashes. By examining variations across influencer types, platforms, and repeated misconduct, this study aims to provide insights into the evolving dynamics of influencer-audience relationships in the context of decentralized finance, with broader implications for policymakers, social media platforms, and influencers.

Understanding the structure and influence of academic journals is essential for evaluating knowledge dissemination and identifying trends across disciplines. This paper introduces a novel approach that combines Loglinear models and Multidimensional Scaling (MDS) to analyze large-scale citation data, leveraging datasets with over 10 million citations. Using advanced web crawling techniques and API-based data extraction from major scholarly databases, we ensure comprehensive and up-to-date data collection across various fields. The Loglinear models capture citation dependencies and trends, while the MDS framework visualizes these relationships in a low-dimensional space for intuitive interpretation. By integrating statistical modeling with visualization, this method provides a robust framework for mapping journals and exploring the dynamics of scholarly communication, offering valuable insights for researchers, institutions, and policymakers.