Multivariate Statistical Analysis
PNNL develops and employs novel data-intensive analytic methods to extract hidden features, anomalies, and signatures from high-dimensional, large-volume, multimedia data in support of discovery and confident decision-making. We develop and apply data mining, clustering, discrimination/classification, process monitoring and control, time series modeling, change detection, pattern recognition, and chemometrics techniques.
We also extract key features from data and track processes and systems while accounting for inherent uncertainties. Feature extraction and signature tracking are applied to large data sets to extract estimates of dynamic physical parameters from waveform data and to facilitate automatic detection of interesting, abrupt changes in waveform data streams. Data forms include text, video, audio, numerical, spectral, categorical, genomic, and imagery.