An Unsupervised Approach to Market Manipulation Detection: Price Anomaly Identification via Isolation Forest

This thesis analyses unsupervised anomaly detection as an approach to uncover possible cases of market manipulation. Detecting this sort of behavior in large financial datasets is not straightforward, and the continuous innovation of financial fraud makes it even more challenging to carry out. Fortunately, Isolation Forest algorithm spots outliers in complex data, using anomalies as premature signs of potential cases of market manipulations. Building a model that could scan price series and highlight unusual movements that don’t fit the typical market pattern was the key objective. One of the advantages of Isolation Forest is that it doesn’t need pre-labeled cases of manipulation to work, it can pick up anomalies on its own. This practically means it can provide a first filter for identifying transactions or price swings, offering a more specific analytic action plan. The final model doesn’t claim to solve market manipulation detection entirely, but it offers a useful starting point. It shows that unsupervised methods can add value as an initial layer of monitoring, supporting more advanced or targeted techniques.