Recent Temporal Patterns for Event Detection in Multivariate Time Series Data

This paper introduces a framework that mines temporal patterns in complex multivariate time series data. Since multivariate temporal time series could be noisy and inaccurately reported, first they are converted into time-interval sequences. Such sequences are then used to build temporal patterns backwards in time using temporal operators, which describe relation between two sequences (before and co-occur). Authors are focused on recent temporal patterns (RTPs), i.e., patterns that can be identified using recent data. RTPs are useful for monitoring and solving event detection problems. Experimental evaluations have been performed on electronic health record data (EHRs), and confirmed efficiency of the newly introduced approach. Precision and recall of generated patterns is computed by using medical diagnosis guidelines to verify patterns.