Time Series Classification Based on Forward Echo State Convolution Network

DOI: https://doi.org/10.1007/s11063-024-11449-8

IF: 2.565

2024-05-12

Neural Processing Letters

Abstract:The Echo state network (ESN) is an efficient recurrent neural network that has achieved good results in time series prediction tasks. Still, its application in time series classification tasks has yet to develop fully. In this study, we work on the time series classification problem based on echo state networks. We propose a new framework called forward echo state convolutional network (FESCN). It consists of two parts, the encoder and the decoder, where the encoder part is composed of a forward topology echo state network (FT-ESN), and the decoder part mainly consists of a convolutional layer and a max-pooling layer. We apply the proposed network framework to the univariate time series dataset UCR and compare it with six traditional methods and four neural network models. The experimental findings demonstrate that FESCN outperforms other methods in terms of overall classification accuracy. Additionally, we investigated the impact of reservoir size on network performance and observed that the optimal classification results were obtained when the reservoir size was set to 32. Finally, we investigated the performance of the network under noise interference, and the results show that FESCN has a more stable network performance compared to EMN (echo memory network).

computer science, artificial intelligence

What problem does this paper attempt to address?

The paper aims to address the problem of time series classification. Specifically, it proposes a new framework called the Forward Echo State Convolutional Network (FESCN). This framework combines the Forward Topology Echo State Network (FT-ESN) with Convolutional Neural Networks (CNN) to improve performance in time series classification tasks.

### Main Contributions:
1. **Proposed Fixed Topology Structure**: To address the issue of random connections in traditional Echo State Networks (ESN), a Forward Topology Echo State Network (FT-ESN) with a fixed forward topology structure is proposed to enhance the stability and classification capability of the network.
2. **Improved Readout Layer**: The traditional linear readout layer is replaced with a new structure that includes convolutional layers and max-pooling layers, namely FESCN.
3. **Experimental Results**: In time series classification tasks conducted on the UCR dataset, FESCN demonstrated overall classification accuracy superior to other methods and showed more stable performance under noise interference.

### Experimental Validation:
- Detailed experiments were conducted on the UCR dataset, comparing FESCN with various traditional methods and deep learning models.
- The results show that FESCN achieved the best performance on multiple datasets and outperformed other methods in comprehensive evaluation metrics such as MPCE, AMR, and GMR.

### Conclusion:
The paper addresses the shortcomings of traditional ESN in terms of network instability and classification performance by proposing FESCN, which shows significant advantages, especially in time series classification tasks.