Distribution skew-based binning: Towards mining highly discriminative patterns from EEG/EMG time series

Abstract

Discovering useful patterns in medical time series data such as EEG and EMG recordings is an important step for gaining useful insights into the data and medical problem under investigation, and for building accurate classifiers. However, pattern mining algorithms often require a binning step, which maps the time series data into a representation in terms of discretized values, in order to discover patterns. How the intervals are constructed has a significant impact on the quality of the mined patterns. We propose a novel binning technique, called Distribution Skew-based Binning (or DS Binning), which uses the distribution of the classes associated with the numerical attribute values to construct the intervals. Experiments show that this method outperforms existing binning methods in facilitating the discovery of high quality patterns from multivariate EEG/EMG time series data, leading to higher classification accuracy. Our experiments demonstrate that DS binning can provide approximately a 5-10% improvement in classification accuracy over other binning methods in multiple scenarios.