Class-specific training improves accuracy of time series classification
Improving Multivariate Time Series Classification with Class-Wise Training and Model Aggregation
Machine Learning
Summary
Classifying multivariate time series data can be tricky because not all parts of the data are equally important for every category. The authors suggest looking at each class separately to find which parts (or dimensions) of the data matter most for that class. They train models dedicated to each class and then combine the results for a final prediction. This method helps reduce noise from less useful data and improves performance, especially when dealing with many dimensions. Their tests show this approach works well across different kinds of datasets and also helps explain which data parts are important for each class.
Multivariate time seriesTime series classificationDimension selectionFeature representationMiniRocketRandom kernelsModel aggregationNoise reductionClass-wise trainingInterpretability
Authors
Mouhamadou Mansour Lo, Gildas Morvan, Mathieu Rossi, Fabrice Morganti, David Mercier
Abstract
In this paper, we propose a class-wise dimension (channel) selection framework for Multivariate Time Series Classification (MTSC). Rather than applying a single global dimension selection process, the proposed approach independently identifies informative dimensions for each class. A dedicated learning process is subsequently performed for each class, followed by a fusion stage for final prediction. The objective is to improve the generation of discriminative feature representations while reducing the influence of noisy or non-informative dimensions. The proposed framework is evaluated using MiniRocket, a random kernel-based baseline method. Experimental results indicate that class-wise dimension selection improves the quality of extracted representations and can enhance classification performance, particularly in high-dimensional settings. These findings suggest that incorporating class-specific information into the training process represents a promising direction for MTSC, improving robustness through consistent gains across heterogeneous datasets, and interpretability through the explicit identification of class-relevant dimensions.