On the Performance of Hierarchical Temporal Memory Predictions of Medical Streams in Real Time

Machine learning is widely used on stored data, recently it is developed to model real time streams. Applying machine learning on medical streams might lead to a breakthrough on emergency and critical care through online predictions. Modeling real time streams implies limitations to the current state-of-the-art of machine learning and requires different learning paradigm. In this paper, we investigate and evaluate two different machine learning paradigms for real time predictions of medical streams. Both the hierarchical temporal memory (HTM) and long short-term memory (LSTM) are employed. The performance assessment using both algorithms is provided in terms of the root mean square error (RMS) and mean absolute percentage error (MAPE). HTM is found advantageous as it provides efficient unsupervised predictions compared to the semi-supervised learning supported by LSTM in terms of the error measures.