A Machine Learning Approach to Detection of Trade-Based Manipulations in Borsa Istanbul

This study investigates trade-based manipulations of capital market instruments. The dataset of the study was gathered from 22 cases of manipulation in Borsa Istanbul (BIST) that occurred in the period between 2010 and 2015. We propose a machine learning approach consisting of supervised machine learning classification models to detect trade-based manipulation from the daily data of manipulated stocks. As a result of this study, supervised machine learning techniques are proven to be successful at detecting trade-based manipulations in trading networks based on the measurement methods of accuracy, sensitivity, and F1 score. We found that our proposed model has an F1 score of 91%, 95% sensitivity, and 93% accuracy in market manipulation detection.