Ambulatory electrocardiograms suffer from muscle and electrode-motion noise that obscures arrhythmia events. We propose a dual-channel adaptive filter that uses accelerometer reference streams synced to a chest patch. On a labelled free-living dataset, the method improves QRS detection sensitivity during walking and stair climbing without attenuating clinically relevant ST-segment morphology in resting segments.

Cyclones strip foliage and fragment mangrove stands, yet recovery trajectories vary with species mix and hydrology. Multi-year Sentinel-2 composites capture greenness rebound across twelve estuaries after a major landfall. Random-forest attribution links faster recovery to proximity of intact seed sources and reduced aquaculture conversion, guiding where assisted regeneration is most likely to succeed.

Student talk that elaborates causal mechanisms may support revision of intuitive misconceptions more than teacher-centred exposition. We code video from twenty lessons on force and motion using an adapted dialogue framework and link turn types to pre–post gains. Multilevel models show exploratory questioning and peer rebuttal predict larger conceptual shifts after controlling for prior achievement, with implications for professional development on orchestrating whole-class discussion.

Experimental screening of glass-transition temperatures remains costly for new monomer libraries. We encode repeat units as attributed molecular graphs and train a message-passing network on a curated dataset of amorphous polymers. Hold-out mean absolute error undercuts group-contribution baselines, and attention maps highlight backbone flexibility and side-chain bulk as chemically plausible drivers of predicted shifts.

Micro and small shops face trade-offs among delivery speed, cost, and reliability when outsourcing restocking. A stated-preference experiment with 412 owners estimates willingness to pay for evening windows and consolidated drops. Latent-class models reveal a cost-sensitive segment and a reliability-focused segment, informing platform pricing and municipal curb-management pilots in dense commercial corridors.

Filter-feeding bivalves integrate particle loads over weeks and can indicate chronic microplastic exposure. Soft tissues from three native mussel species collected along a river transect show polymer fragment counts increasing toward combined sewer outfalls. FTIR typing links dominant polymers to packaging and textile fibres, supporting targeted source-control measures upstream of drinking-water intakes.

Pooling patient-level records across institutions improves predictive accuracy but raises privacy and governance barriers. We train a federated logistic and gradient-boosted ensemble on de-identified stroke admissions from five regional hospitals without centralizing raw data. Cross-site validation shows discrimination comparable to a pooled oracle model while satisfying local data-retention policies, and ablation of feature sets highlights blood-pressure trajectories as the dominant shared signal.