Enhanced Predictive Modelling for 30-Day Readmission Diabetes Patients Based on Data Normalization Analysis

Abstract

In the field of health care, one of the most important problems is predicting the possibility of hospital readmission due to its important role in caring for patients with chronic diseases such as diabetes. Such predictions affect the health care costs and the hospital’s efficiency and reputation. In this paper, an intelligent-based model is developed to predict the reintroduction of the patient into the hospital. This model is based on using some Machine Learning (ML) algorithms such as Logistic Regression (LR), K-Nearest Neighbors (KNN), and Support Vector Machine (SVM). Also, it proposes the use of a Deep Learning (DL) based network such as a convolutional neural network (CNN). Both ML and DL are used as classifiers to predict hospital readmission. The main problem is the input noisy data to these classifiers. These noisy data reduce the accuracy of the readmission prediction model. Sequential pre-processing steps are proposed to get over such a problem. These pre-processing steps provide solutions to missing values, feature engineering, and normalization problems. The main contribution of this work is improving readmission prediction rate by solving the data normalization problem. Two types of data normalization (e.g. z-score and min-max normalization) are applied, results show there is a difference in accuracy, z-score normalization is better than min-max normalization when comparing ML methods and DL models, CNN is the best with an accuracy of 0.894% in case of z-score normalization. Moreover, the model performance is improved with an accuracy of 0.924% when non-normalized data is used as input to the model. The proposed Non-normalization technique successes in providing superior results compared to some previous techniques which are displayed data by using Ensemble, Normalization, and Ensemble by age group techniques.