Time Series Classification Based on FCN Multi-scale Feature Eensemble Learning

Time series classification problem (TSC) is of great significance in the fields of finance, health care and environment. The full convolutional network is prominent in the classification of time series, but the feature length of the original data extracted by the network is fixed. Since the highly distinguishable local features in time series may have different scales, this paper proposes a multi-scale feature ensemble full convolutional network (MFCN) to extract different scale features and improve the classification accuracy of the network. Finally, 11 excellent time series classification models were compared on 44 UCR data sets, and the results showed that the proposed method obtained the minimum classification accuracy error in 16 data sets, and the average accuracy was also the highest, reaching 87.7%.