Tamara Sipes

Large Scale predictive analytics for real-time energy management

As demand for cost-effective energy and resource management continues to grow, intelligent automated building solutions are necessary to reduce energy consumption, increase alternative energy sources, reduce operational costs and find interoperable solutions that integrate with legacy equipment without massive investments in new equipment and tools. The ability to analyze, understand and predict building behavior offer tremendous opportunities to demonstrate and validate increased energy efficiencies, which may ease many particular exorbitant pressures taxing the grid. In this paper, we describe a research platform driven by an existing campus microgrid for developing large scale, predictive analytics for real-time energy management.

A MULTIVARIATE TIME SERIES CLASSIFICATION METHOD FOR STREAMING DATA USING TEMPORAL METAFEATURE ABSTRACTIONS

In this paper we demonstrate a new approach to the classification of multivariate time series streaming data by utilizing a temporal metafeature abstractions method. The technique extracts global features and metafeatures in order to capture the necessary time-lapse information in the streams of data. The features are then used to create a static, intermediate stream representation that includes all the important time-varying information, and is suitable for analysis using the standard supervised data mining techniques. The capability of the new algorithm called MineTool-TS2 was demonstrated through its application to three datasets: UCSD Microgrid energy usage data, a space physics dataset and synthetic data.

Anomaly Detection in Healthcare: Detecting Erroneous Treatment Plans in Time Series Radiotherapy Data

Adverse events in healthcare and medical errors result in thousands of accidental deaths and over one million excess injuries each year. Anomaly detection in medicine is an important task, especially in the area of radiation oncology where errors are very rare, but can be extremely dangerous, and even deadly. To avoid medical errors in radiation cancer treatment, careful attention needs to be made to ensure accurate implementation of the intended treatment plan. In this paper, we describe the work that resulted in a valuable predictive analytics tool for automatic detection of catastrophic errors in cancer radiotherapy, adding an important safeguard for patient safety. We designed a method for Dynamic Modeling and Prediction of Radiotherapy Treatment Deviations from Intended Plans (SmartTool) to automatically detect and highlight potential errors in a radiotherapy treatment plan, based on the data from several thousand prostate cancer treatments that were used to build the model. SmartTool determines if the treatment parameters are valid, against a previously built Predictive Model of a Medical Error (PMME). SmartTool communicates with a radiotherapy treatment management system, checking all the treatment parameters in the background prior to execution, and after the human expert QA is completed. Any anomalous treatment parameters are detected using an innovative intelligent algorithm in a completely automatic and unsupervised manner, and it flags the operator by highlighting the suspect parameter(s) for human intervention. Furthermore, the system is self-learning and constantly evolving, and the model is dynamically updated with the new treatment data.

Anomaly Detection in Time Series Radiotherapy Treatment Data

The work presented here resulted in a valuable innovative technology tool for automatic detection of catastrophic errors in cancer radiotherapy, adding an important safeguard for patient safety. We designed a tool for Dynamic Modeling and Prediction of Radiotherapy Treatment Deviations from Intended Plans (Smart Tool) to automatically detect and highlight potential errors in a radiotherapy treatment plan, based on the data from several thousand prostate cancer treatments at Moore Cancer Research Center at University of California San Diego. Smart Tool determines if the treatment parameters are valid, against a previously built Predictive Model of a Medical Error (PMME). Smart Tool has the following main features: 1) It communicates with a radiotherapy treatment management system, checking all the treatment parameters in the background prior to execution, and after the human expert QA is completed, 2) The anomalous treatment parameters, if any, are detected using an innovative intelligent algorithm in a completely automatic and unsupervised manner, 3) It is a self-learning and constantly evolving system, the model is dynamically updated with the new treatment data, 4) It incorporates expert knowledge through the feedback loop of the dynamic process which updates the model with any new false positives (FP) and false negatives (FN), 4) When an outlier treatment parameter is detected, Smart Tool works by preventing the plan execution and highlighting the parameter for human intervention, 5) It is aimed at catastrophic errors, not small errors.