We Need a More Robust Classifier: Dual Causal Learning Empowers Domain-Incremental Time Series Classification

TL;DR

This work tackles domain-incremental time series classification (DI-TSC), where models must learn from streaming domains without forgetting previously seen ones. It introduces DualCD, a lightweight framework that first disentangles temporal representations into class-causal ($\mathbf{Z}_R$) and spurious ($\mathbf{Z}_I$) components using orthogonal masks, then enforces robustness through a dual causal intervention that generates intra-class and inter-class variant samples. The model optimizes a pair of causal-intervention losses to ensure predictions rely on causal features, and it is validated across four real-world datasets and multiple backbone models, achieving state-of-the-art results with strong evidence of reduced forgetting. The work also establishes a DI-TSC benchmark and a new PRF metric to better quantify knowledge consolidation in domain-incremental settings, highlighting DualCD's practical potential for robust, privacy-preserving continual learning in time series tasks.

Abstract

The World Wide Web thrives on intelligent services that rely on accurate time series classification, which has recently witnessed significant progress driven by advances in deep learning. However, existing studies face challenges in domain incremental learning. In this paper, we propose a lightweight and robust dual-causal disentanglement framework (DualCD) to enhance the robustness of models under domain incremental scenarios, which can be seamlessly integrated into time series classification models. Specifically, DualCD first introduces a temporal feature disentanglement module to capture class-causal features and spurious features. The causal features can offer sufficient predictive power to support the classifier in domain incremental learning settings. To accurately capture these causal features, we further design a dual-causal intervention mechanism to eliminate the influence of both intra-class and inter-class confounding features. This mechanism constructs variant samples by combining the current class's causal features with intra-class spurious features and with causal features from other classes. The causal intervention loss encourages the model to accurately predict the labels of these variant samples based solely on the causal features. Extensive experiments on multiple datasets and models demonstrate that DualCD effectively improves performance in domain incremental scenarios. We summarize our rich experiments into a comprehensive benchmark to facilitate research in domain incremental time series classification.

Figures (5)

• Figure 1: The framework of DualCD, which includes a temporal disentanglement module and a dual causal intervention mechanism, aims to learn class-invariant causal features.
• Figure 2: Ablation study on the four datasets.
• Figure 3: Hyperparameter analysis on the HAR dataset.
• Figure 4: Visualization analysis conducted on the test set of domain 1 from the HAR dataset.
• Figure 5: Distributional differences in representations of the first class across ten domains on the HAR dataset.