Multi-Source Change-Point Detection over Local Observation Models

• Bayesian change-point detection on multi-source temporal sequences of observations with mixed statistical type and high dimension Local observation models for support alignment adaptive methodology Explainability the contribution each source to global Real-world experiments smartphone-based monitoring dataset healthcare In this work we address problem (CPD) high-dimensional, multi-source, heterogeneous sequential data missing values. We present a new CPD based local latent variable factorizations that enhances fusion different data-type face dimensionality. Our motivation comes from behavioral change in measured by smartphone monitored Electronic Health Records. Due differences relevance information, other works fail obtaining reliable estimates change-points location. This leads methods are not sensitive enough when dealing interspersed changes intensity within same sequence or partial components. Through definition (LOMs), transfer CP information homogeneous spaces propose several weight CPD. With presented demonstrate reduction both delay number not-detected CPs, together robustness against presence values synthetic dataset. illustrate its application real-world study add explainability degree contributing detection.