Development and Research of Methods For Detecting Market Deal Anomalies

Market price manipulation is an illegal practice. This work aims at developing methods for detection of market price manipulations giving high accuracy with minimal number of errors of the first kind and errors of the second kind. To achieve this goal the author analyzed the existing price manipulations on the different kind of exchanges, analyzed the existing data anomaly detection methods, analyzed the existing approaches to the price manipulation detection in the market data of different kind, analyzed the given data sets and applied machine learning algorithms to the data sets, optimized the algorithms, and analyzed the results. This work treats price manipulations as market data anomalies. The data were aggregated for some time intervals by the selected features. The following classification algorithms were applied to the tagged data: Random Forest, Decision Tree, KNN Classifier and Naive Bayes Classifier. They were tuned and optimized with respect to the data. The Random Forest algorithm achieved the best result with the minimum number of false positives. The Isolation Forest algorithm was applied to the set of unlabeled data and according to expert assessment also showed a good result.