DATA SCIENCE-DRIVEN APPROACHES TO IDENTIFYING AI-GENERATED CONTENT: MACHINE LEARNING AND NLP MODELS FOR ACADEMIC INTEGRITY AND DIGITAL TRANSPARENCY

Abstract

In the rapidly evolving digital landscape, the prevalence of generative artificial intelligence (GenAI) systems and large-language models (LLMs) has created profound challenges for academic integrity and content authenticity. This paper proposes a data science-driven framework for identifying AI-generated content in academic and digital environments by leveraging advanced machine learning (ML) techniques and natural language processing (NLP) models. First, we survey the current state of AI-generated content detection, reviewing both traditional ML approaches and state-of-the-art transformer-based architectures, and we demonstrate that while recent systems can achieve high accuracy (e.g., over 90 %) in controlled settings, significant limitations remain—especially regarding fairness, generalisability, and bias against non-native English writers. Next, we develop and evaluate a hybrid detection model that combines feature-engineering (lexical, syntactic, stylometric) with embedding-based representations and a supervised classifier trained on a curated dataset of human-written versus AI-generated academic prose. We integrate explainable-AI (XAI) techniques to interpret model decisions and identify the most discriminative features distinguishing human and machine authorship. Our results indicate that the proposed model outperforms baseline detectors in both accuracy and transparency, and we further examine its application to institutional workflows for academic integrity, such as submission screening and authenticity audits. Finally, we discuss ethical, operational and policy implications of deploying such detection systems in higher-education settings, including issues of false-positives, equity, transparency and the evolving “arms-race” between AI generation and detection. By framing detection as part of a broader ecosystem of digital transparency and trust, this research contributes both methodologically and practically to safeguarding academic standards in the era of AI-augmented writing.