GRU Deep Residual Network for Time Series Classification
Fuyu Zhu, Hua Wang, Yixuan Zhang · 2023
In this paper, we propose a simple but powerful model for time series classification with deep neural networks. The proposed model is purely end-to-end, without any heavy pre- processing or feature production on the raw data. The proposed Gated Recurrent Residual Full Convolutional Network (GRU- ResFCN) achieves superior performance compared to other state- of-the-art approaches.GRU-ResFCN combines a gated recurrent network GRU module to extract temporal features of the data and a residual network neural network (ResFCN) module to efficiently and quickly extract null domain features, using an attention mechanism to improve time series classification. Our exploration of very deep neural networks using ResNet structures is competitive. The global average pool in the convolutional model is able to use class activation maps to identify regions in the original data that contribute to specific labels. The model provides a simple alternative for real-world applications and a good starting point for future research. We also provide a comprehensive analysis of the model’s generalization capability, learned features, network structure, and classification semantics.