DTWSSE: Data Augmentation with a Siamese Encoder for Time Series

Access to labeled time series data is often limited in the real world, which constrains performance of deep learning models field analysis. Data augmentation an effective way solve problem small sample size and imbalance datasets. The two key factors are distance metric choice interpolation method. SMOTE does not perform well on because it uses a Euclidean interpolates directly object. Therefore, we propose DTW-based synthetic minority oversampling technique using siamese encoder for named DTWSSE. In order reasonably measure series, DTW, has been verified be method forts, employed as metric. To adapt DTW metric, use autoencoder trained unsupervised self-training manner interpolation. Siamese Neural Network mapping from hidden space feature space, decoder used map back space. We validate proposed methods number different balanced or unbalanced Experimental results show that can lead better downstream model.