Neural Guidance for Predicting Early Readmission in Diabetic Patients

Abstract

This study builds on the first paper's look at using machine learning to predict which diabetic patients will need to go back to the hospital. The information used comes from Kaggle and is made up of details of diabetic patients. The first steps include importing packages, exploring datasets, and cleaning, which includes changing binary classes and getting rid of columns that aren't needed. Using Seaborn and Matplotlib to make a full picture of the information helps people understand it better. LabelEncoder is used for label encoding, and feature selection methods are used. After that, the data is split into sets that are used to train and test both deep learning and machine learning models. As part of the study, different models for binary and multi-class classification were built. These models include S V M, Random Forest, Guided Artificial Neural Networks (ANN) with different optimization methods, and Voting Classifier and Stacking Classifier ensemble methods. It was also very accurate when a combination of Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) was used. The study also includes a Flask framework that is connected to SQLite for user registration and lets users enter feature values for forecast. The data that has already been handled is then used by the learned models, and the end results are shown on the front end. The study was expanded by using ensemble methods, which showed that CNN + LSTM is more accurate than past methods. Adding user identification and interface development makes the models more useful in real life, giving us a more complete way to predict diabetic return.