Disease Prediction Using Electronic Health Records with Long Short-Term Memory Networks

Abstract

With the widespread application of electronic health records (EHR), data-driven disease prediction has become an important research direction in the medical field. This study proposes a disease prediction model based on long short-term memory network (LSTM) to analyze time series data in electronic health records and predict patients' future disease risks. As a deep learning model with long-term dependency modeling capabilities, LSTM can effectively process complex time series features in electronic health data. We used the public MIMIC-III database, which contains a large number of patients' diagnosis, treatment and physiological data, and built a disease prediction system through data preprocessing, feature selection and model training. Experimental results show that LSTM shows superior performance in evaluation indicators such as mean square error (MSE), root mean square error (RMSE) and mean absolute error (MAE) compared with traditional machine learning models such as support vector machine (SVM), random forest (RF) and multi-layer perceptron (MLP). By further optimizing the LSTM model, the accuracy of disease prediction can be improved, providing clinicians with a scientific and reliable auxiliary decision-making tool.