A ResNet with GA-based Structure Optimization for Robust Time Series Classification

Time Series Classification (TSC) received wide attention in machine learning and data mining, and arise in a wide range of fields, such as scheduling, logistics, medical and health etc. How to overcome the noise in the time series datasets is one of the key challenges of the TSC. In this paper, we propose a hybrid Residual Network (ResNet) with a genetic algorithm-based network structure optimization for robust TSC, which is named as GA-ResNet. Although network structure optimization is one of the key ways to obtain an effective deep neural network model, but this structure optimization is a NP-hard problem. We design a genetic algorithm for the ResNet structure optimization. Several benchmarks are adopted to prove the effectiveness of proposed GA-ResNet compared with six state-of-the-art deep neural networks.