Detect temporal misalignment in sensor data to improve classification
Diagnosing Temporal Misalignment in Multichannel Time-Series Classification with Minimum Description Length
Machine LearningArtificial IntelligenceHardware Architecture
Summary
When multiple sensors record time-based data, they are usually assumed to be perfectly in sync. However, small delays or timing issues can happen, hurting the performance of computer models that analyze the data. The authors propose a way to detect these timing mismatches without needing extra labels or retraining models. By shifting the data streams and measuring how well one sensor’s data can describe another’s, their method finds the best alignment and reveals hidden delays. This can help fix timing errors that standard evaluations might miss.
What this means in practice
- •For industrial sensor teams: Identify and correct hidden timing offsets in collected sensor data to improve machine learning model accuracy during deployment.
- •For wearable device developers: Audit and adjust time delays between wearable sensors without retraining models to maintain reliable activity recognition under real-world conditions.
Authors
Sebastian Buschjäger, Michael Frichert, Daniel Kuhe, Jian-Jia Chen
Abstract
Multichannel time-series classification commonly assumes synchronized sensor streams, although latency, clock drift, and preprocessing can introduce relative delays during data collection or after deployment. Existing synchronization solutions are often hardware-specific and difficult to apply retrospectively. Consequently, synchronization problems may remain undetected while classification performance is suboptimal. We introduce a classifier- and label-free diagnostic based on minimum description length (MDL). Our method applies candidate temporal shifts to sensor groups and measures how efficiently one group can be encoded through a representation of the remaining channels. An increased codelength indicates that the shift destroys shared temporal structure, whereas the minimum identifies the alignment most strongly supported by the data. Unlike learned synchronization methods, the diagnostic requires neither retraining nor a trusted aligned reference and can therefore test both training and deployment data for misalignments. Experiments on two controlled synthetic tasks and nine real-world datasets show that the metric exposes alignment structure and can recover accuracy under induced deployment drift. A whole-dataset audit further identifies stable nonzero MDL optima in established benchmarks including FordChallenge, Opportunity, PAMAP2, and UCIActivity, revealing potential systematic offsets that conventional model evaluation does not expose. Our method thus provides a general-purpose tool for detecting, understanding, and correcting temporal misalignment throughout the time-series learning pipeline. Our code is available under https://github.com/sbuschjaeger/mdl-temporal-misalignment.