Prediction and Detection of Anomalous Events

Prediction of anomalous or rare events, such as the automated identification of fraudulent financial transactions, the prediction of novel microRNA molecules within the human genome, or the correct diagnosis of rare cancerous biopsies, is both important and challenging. This lab focuses on the development and evaluation of statistical models and machine learning systems for mixed mode data, and on making predictions in the presence of class imbalance. Graduate training is critically needed in this area since many events of interest in the real world tend to be rare.

Our lab offers training to address the unique challenges associated with applying statistical machine learning for the prediction of rare events. The inherently low prevalence of rare events must be explicitly addressed during both the training and testing of such systems; otherwise systems either dramatically under- or over-predict rare events. Application and extension of software such as HDoutliers in R is part our research focus. Application areas include biomedical informatics, network security applications, sports analytics, and business informatics.