分析u-Health系统心电图数据并进行早期分类，可给医生提供紧急警报以便及时治疗甚至挽救生命。本项目拟基于时间序列shapelets对心电图数据提取特征，利用提取的特征实现可解释性的早期分类。心电图的连续监控数据以多元时间序列的形式存在，时间序列shapelets是时间序列中最具有辨别性的子序列，具有可解释性，而多视图数据可以获得比单视图数据更高的模型性能。拟研究多元时间序列中的局部shapelets；基于multi-view的观点对时间序列的shapelets特征和心电图的波形特征做集成特征选择；基于集成的特征设计分类器提前分类。综合上述研究，以期实现心电图数据的可靠分类,确保分类的早期性、准确率和可解释性，可望在理论上丰富和拓展时间序列数据挖掘的内涵,在实践上应用于实际医疗健康领域。

Analyzing electrocardiogram data in u-Health system and making early classification can provide emergency alert to doctor so as to make timely medication or surgery even save time to save lives. This project will extract features in ECG data based on time series shapelets, and utilize the features to realize interpretable early classification.Continuous monitoring ECG data mainly exist in the form of multivariate time series, while shapelets are discriminative subsequences in time series with high interpretability, and multi-view data can obtain better model performance than single-view data. The project will study deeply from the following aspects: local shapelets discovery of multivariate time series,multi-view feature ensemble on time series shapelets and ECG wave feature, and then design classifiers to do early prediction. Through this research, it is expected to realize reliable classification of ECG data and ensure the earliness,accuracy and interpretability of classification,theoretically enrich and expand the connotation of time series data mining and practically be applied to the actual healthcare field.