Machine Learning-Driven Early Warning Analytics for Identifying Market Manipulation, Irregular Trading Activity, and Suspicious Market Signals in U.S. Stock Markets

Abstract

This paper develops a machine learning-driven early warning framework for identifying market manipulation, irregular trading activity, and suspicious market signals in U.S. stock markets, while grounding the argument in public regulatory evidence rather than purely illustrative examples. The study is motivated by three practical problems. First, conventional rule-based surveillance remains necessary but often produces large alert volumes, high false-positive burdens, and limited capacity to prioritize weak but meaningful signals. Second, suspicious trading increasingly spans multiple channels, including order flow, cross-venue behavior, public narratives, and relational patterns that are poorly captured by isolated tabular indicators. Third, the strongest public evidence available to researchers is fragmented across enforcement summaries, disciplinary statistics, and whistleblower reporting, which means rigorous early warning research must integrate official sources even when transaction-level labels remain limited. Accordingly, the manuscript combines scholarly synthesis with descriptive evidence derived from Securities and Exchange Commission and Financial Industry Regulatory Authority datasets and reports covering the 2021-2023 period.