Contextual LSTM (CLSTM) Models for Early Prediction of Sepsis

Chen Yao
Sichuan University


Abstract

Early detection of sepsis is critical for improving sepsis outcomes. In this challenge, prediction of sepsis can be regarded as sequence tagging task. We present CLSTM (Contextual LSTM) for solve the problem. CLSTM is an extension of the recurrent neural network LSTM (Long-Short Term Memory) model, where we incorporate contextual features into the model. Contextual features are used in order to reduce overfitting and improve model performance. The great emphasis is put on the method of dynamic data sampling to represent the characteristics of data, this method can generate a large amount of extratraining data segmenting. These measures improve robustness and generalization ability.