Clustering Framework Detects Suspicious Trading Patterns in Capital Markets

Researchers have developed an unsupervised fraud-detection framework aimed at identifying suspicious trading patterns within capital markets. The toolkit begins with K-Means++ clustering, applied to a substantial dataset of approximately one million financial transactions spanning over a decade. The primary goal is to combat market manipulation, which erodes confidence in trading platforms. The proposed methodology highlights 2.02% of the analyzed trades as suspicious. Further categorization, based on market practice heuristic thresholds, revealed that 51.10% of these suspicious trades indicated spoofing, 0.10% suggested pump and dump schemes, 0.55% pointed to insider trading, and 1.43% indicated fake breakouts. A significant portion (46.83%) remained unclassified. Despite the absence of ground truth labels, the model's performance was supported by a Silhouette Score of 0.561, indicating reasonable clustering quality.