Multivariate Time Series Classification Based on Multi-Channel Representation Model and Broad Learning System

Multivariate time series classification is a crucial topic in data mining with far-reaching implications in areas such as finance, healthcare, and industry. In this paper, a novel multivariate time series classification method is proposed. First, a representation model based on local, global, and time-dependent information is constructed. In order to effectively supervise training process of the representation model, an adaptive joint loss function is designed so that the weight of the loss module can be dynamically adjusted. Subsequently, the trained representation model is employed to map original data into a new representation space. In addition, a broad learning system is applied to classify these representations. According to some experimental results on 30 benchmark datasets, the proposed method can achieve better performance in multivariate time series classification tasks.