AdaFSNet: Time Series Classification Based on Convolutional Network with a Adaptive and Effective Kernel Size Configuration

Haoxiao Wang,Bo Peng,Jianhua Zhang,Xu Cheng

2024-04-29

Abstract:Time series classification is one of the most critical and challenging problems in data mining, existing widely in various fields and holding significant research importance. Despite extensive research and notable achievements with successful real-world applications, addressing the challenge of capturing the appropriate receptive field (RF) size from one-dimensional or multi-dimensional time series of varying lengths remains a persistent issue, which greatly impacts performance and varies considerably across different datasets. In this paper, we propose an Adaptive and Effective Full-Scope Convolutional Neural Network (AdaFSNet) to enhance the accuracy of time series classification. This network includes two Dense Blocks. Particularly, it can dynamically choose a range of kernel sizes that effectively encompass the optimal RF size for various datasets by incorporating multiple prime numbers corresponding to the time series length. We also design a TargetDrop block, which can reduce redundancy while extracting a more effective RF. To assess the effectiveness of the AdaFSNet network, comprehensive experiments were conducted using the UCR and UEA datasets, which include one-dimensional and multi-dimensional time series data, respectively. Our model surpassed baseline models in terms of classification accuracy, underscoring the AdaFSNet network's efficiency and effectiveness in handling time series classification tasks.

Machine Learning,Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The paper aims to address a key challenge in Time Series Classification (TSC), which is how to capture the appropriate Receptive Field (RF) size from one-dimensional or multi-dimensional time series data of different lengths. Although time series classification has wide applications in various fields such as finance, environmental research, and healthcare, selecting the appropriate time scale to extract features remains a significant challenge. Traditional machine learning methods and deep learning methods are both attempting to solve this problem, but as the length of time series data increases, the demand for computational resources rises sharply, posing a huge challenge for model optimization.

To address this issue, the authors propose a method called AdaFSNet (Adaptive Full-Scale Convolutional Neural Network), which can dynamically select a range of kernel sizes to effectively cover the optimal receptive field sizes for different datasets. Specifically, AdaFSNet includes two dense blocks and designs a TargetDrop module to reduce redundancy while extracting more effective receptive fields. Experiments on the UCR and UEA datasets show that AdaFSNet performs excellently in terms of network training speed and convergence, and surpasses baseline models in classification accuracy, demonstrating its effectiveness and efficiency in handling time series classification tasks.