Rahul Bordoloi (Univ. of Rostock) : Multivariate Functional Linear Discriminant Analysis of Partially-Observed Time Series

As access to biomedical time-series data expands, new challenges arise—particularly in handling missing values and irregular sampling. Traditional methods like Linear Discriminant Analysis (LDA) are not well-suited to this kind of fragmented data. In this talk, we introduce MUDRA, a multivariate extension of Functional Linear Discriminant Analysis (FLDA), designed to address two core issues: (1) statistical dependencies across features of a multivariate time series, and (2) missing or irregularly sampled features. MUDRA uses a computationally efficient Expectation/Conditional-Maximization (ECM) algorithm that avoids expensive tensor operations and handles missing data without imputation or padding. This approach offers a principled, scalable solution for classification and dimension reduction in complex biomedical time-series datasets.