Prediction of Sepsis Using LSTM Neural Network With Hyperparameter Optimization With a Genetic Algorithm

Here in this study, we propose an algorithm for the prediction of sepsis using Long Short-Term Memory (LSTM) neural networks. The algorithm was trained and validated on public dataset proposed by PhysioNet Challenge 2019. We have used a differential evolution genetic algorithm to find optimal training hyperparameters and probability threshold for the inference phase. The deployed version of the algorithm obtained normalized utility score 0.278 based on full challenge test set (team name: ISIBrno; rank 27 of 78). Presented method required 50 mins of single-core CPU processing time.