Increasingly large volumes of time series medical data need to be exploited to learn better models that forecast events such as acute hypotensive episodes (AHE). The models have to be transparent so clinicians can decipher and validate them with their expert knowledge. The nature of learning a forecasting model requires historical ”lag” data but in most cases the extent of the relevant lag is open to question and the cost, with respect to model learning, increases with its duration. We present a novel decision list-based machine learning approach for forecasting physiological time series by classification. It is scalable, finds lag duration automatically and the rules it learns are interpretable and compact in terms of their representation of lagged variables.

PDF

In Workshop on Data Mining for Medicine and Healthcare at the 14th SIAM International Conference on Data Mining, 82-87, 2014.

@inproceedings{hemberg:siamworkshop14, title={Learning Decision Lists with Lags for Physiological Time Series}, author={Erik Hemberg and Kalyan Veeramachaneni and Prashan Wanigasekara and Hormoz Shahrzad and Babak Hodjat and Una-May O'Reilly}, booktitle={Workshop on Data Mining for Medicine and Healthcare at the 14th SIAM International Conference on Data Mining}, month={ }, pages={82-87}, url="http://nn.cs.utexas.edu/?hemberg:siamworkshop14", year={2014} }