A study of kNN using ICU multivariate time series data
Admir Djulovic · 2014
A time series is a sequence of data collected at successive time points. While most techniques for time series analysis have been focused on univariate time series data at fixed intervals, there are many applications where time series data are collected at irregular and uncertain time intervals across multiple input variables. The uncertainty in multivariate time series makes analysis difficult and challenging. In this research, we study kNN classification approach applied to ICU multivariate time series data for patient’s mortality prediction. We propose three time series representation strategies to handle irregular multivariate time series data. The experiments show the performance of these three methods in different settings. We also discuss the impact of imbalanced class distribution and the effect of k in kNN classification.