Multivariate Time-series Shapelets And Its Application In ICU Medical Application

Time series data is a series of data with timestamps that change over time.Time series data are generally domain-specific,and have high-dimensional nature,which led to the traditional data mining methods cannot efficiently deal with such data and cannot get useful knowledge.Time series classification problem is a kind of classical problem in time series data mining.The existing classification method of time series classification can be poorly interpreted and the classification speed is not excellent.Time series shapelets is a subsequence with high degree of discernibility of sequence samples and can fastly and accurately classify time series data.ICU medical prediction is a typical multivariate time series classification problem,its accurate prediction has great significance.Therefore,this paper starts the following research based on the above situations:Firstly,a multivariate time series shapelets model is constructed.The dimension of multivariate time series data is high,and its corresponding multivariate time series shapelets model can be established based on the idea of statistical distribution of points.The basic idea is following: local feature point set is extracted after normalization for all the attributes of multivariate time series data;quantiles are extracted according to the idea of that the quantiles can fully describe the distribution of local feature point set;shapelets candidate set is constituted by subsequences of sequence with quantiles and then is screened for optimal k global shapelets by global shapelets extraction method(using the similarity between the multivariate time series and whether the information gain value is impro ved);finally,these optimal k global shapelets are converted into the new shapelets dataset according to the given distance metric.Secondly,an ICU patient death prediction framework based on multivariate time series shapelets is proposed.In the shapelets construction phase,the physiological index matrix model of any patient in ICU is obtained after missing value processing for ICU dataset.The optimal k global shapelets are extracted according to the multivariate time series shapelets model and the new data sets are constructed.In the forecasting phase,normalization for all the attributes of multivariate time series data of new patient and the classifier is constructed by using the data set constructed in the previous stage,the optimal classifier is used to classify the multivariate time series of the new patient and finally the classify result is obtained.Thirdly,the effectiveness of the algorithm and the practicability of the algorithm framework are demonstrated through the experimental results.