Research On ICU Medical Prediction Method Based On Time Series Medical Data

The ICU medical prediction is to determine the physical condition and the trend of the disease development by studying and analyzing the patient's clinical medical data.A good predictive model can help doctors develop better medical plans and avoid over-treatment and treatment delays.ICU medical prediction always relies on large-scale medical datasets.Due to the single access to medical data,the existing medical datasets have problems with sample classes imbalance,which seriously affect the analysis of medical prediction models.In the aspect of prediction model construction,the traditional ICU medical prediction model can not effectively perceive the change of local state of time series medical data,so the utilization of time series data is limited,and the clinical value is small.Therefore,it is of great significance to study the ICU medical prediction method based on time series medical data.In view of the above problems,this paper proposes a classes balance method of time series data sample based on FastDTW and a refused Bi-LSTM medical prediction model based on temporal state monitoring mechanism.The main tasks are as follows:1?In order to solve the problem of sample classes imbalance in medical dataset and improve the quality of time series medical dataset,this paper proposes a classes balance method of time series data sample based on FastDTW,which effectively improves the sample construction efficiency and balance effect.2?In order to improve the performance of medical prediction model,this paper studies the construction methods of multiple medical classification prediction models based on CNN and LSTM.Aiming at the local characteristics of data dependence in medical prediction,this paper proposes a construction method of prediction model based on time series state supervision mechanism,which effectively enhances the ability of prediction model to capture local state changes of time series indicators.And the convergence speed of the model is also improved;For the influence of unknown classes samples on the classification accuracy of labels,we extended the traditional classification problem to the refused classification problem,and proposed the construction method of the refused classification model,which improves the analytical results of medical prediction models.3?Finally,with the experiment analysis,it is proved that the classes balance method of time series data sample based on FastDTW and the refused medical prediction model based on the temporal state monitoring mechanism are effective.