Invariant Subspace Learning for Time Series Data Based on Dynamic Time Warping Distance

•A novel invariant subspace learning framework under DTW is proposed, which jointly solves the subspace learning and the alignment of multiple sequence samples. To the best of our knowledge, this is one of the first algorithms to explore the intrinsic subspace based on DTW distance, instead of Euclidean distance.•The mutual promotion relationship between multiple sequence alignment and subspace learning is investigated and discussed.•Experimental results show that the proposed subspace representation outperforms the state-of-the-art distance-based and feature-based methods in classification tasks.