the Artificial intelligence (AI) has gradually changed from frontier technology to practical application with the continuous progress of deep leing technology in recent years.In this paper,the Random forest (RF) algorithm is adopted to preprocess and optimize the feature subset of ICU data sets.Then these optimized feature subsets are used as input of Long shortterm memory (LSTM) deep leing model,and the early disease prediction of ICU inpatients is carried out by the method of neural network deep leing.Experiments show that this prediction method has higher prediction accuracy compared with other machine leing and deep leing models.