Domain Adaptation of Time Series Classification

In the era of big data and the rapid development of sensor technology, the time series classification problem has become an important research direction in the field of data mining. The complexity and diversity of time series data itself may lead to the inconsistency of the distribution between the training set and the test set, and this distribution difference may cause significant fluctuations in the model performance in practical applications. In contrast, domain adaptation models aim to adapt to test data with different distributions from the training data to improve the model performance. To address this challenge, we propose the Au-RAINCOAT model based on the RAINCOAT model, which is a closed-set domain adaptation model for complex time series. The model first considers the introduction of an Auto-Correlation mechanism module, which aims to compensate for the shortcomings of the Fourier transform by capturing local information while improving the computational efficiency. Second, to avoid the stability problems associated with the choice of regularization parameters in Sinkhorn divergence, we choose to use the Proxy-A-Distance(PAD).In this paper, we use four common datasets such as WISDM, HAR, HHAR, Sleep-EDF, and experiments are conducted on seven models. The experimental results show that the method has better results on the four datasets.