Contrastive Learning-Based Time Series Classification in Healthcare

With the rapid increase in the global elderly population, the shortage of professional care institutions has become increasingly prominent. Against the backdrop of rapid advancements in artificial intelligence technology, caregiving robots have emerged as an innovative solution to alleviate this crisis. This study introduces a novel contrastive learning model specifically called CL-TCH designed for handling time series data related to healthcare. In this model, various data augmentation methods are employed to create positive and negative pairs. The input data is encoded using a Transformer encoder to comprehensively capture features. During the model training process, losses are calculated in both temporal and spatial dimensions. The model is validated on three public datasets, and three ablation experiments are conducted to demonstrate the necessity of each module. Experimental results show that our approach exhibits excellent performance in tasks related to time series classification in the context of healthcare.