Stock Price Manipulation Detection Using Empirical Mode Decomposition Based Kernel Density Estimation Clustering Method

Stock market manipulation means illegitimate or illegal activities trying to influence the prices of stocks, hence diluting the legal definition of trading stocks. In this research, a model for detecting Stock price manipulation is presented for anomalies like Pump & Dump, Quote stuffing, Gouging or Spoof Trading. The model is presented on level 1-tick data which contains highly volatile time series and a high trading frequency making the detection more challenging. In literature, very less number of studies based on unsupervised learning for Stock market manipulation has been carried out. In addition, the existing studies focused only on specific anomalies and were not generalized enough to capture other anomalies. The research model used in this work uses unsupervised learning where the input data is decomposed using Empirical Mode Decomposition followed by Kernel Density Estimation based clustering technique for anomaly detection. One of the key advantages of this technique is receiver operating characteristic (ROC) curve, which is better than the currently available techniques and provides a maximum area under curve (AUC) equal to 0.96. The results in this work are also compared with existing benchmark approaches like K-means, Principal Component Analysis (PCA) based anomaly detection and Dirichlet process Gaussian Mixture Model (DPGMM) based anomaly detection, and a maximum improvement of 84% is obtained.