Forecasting Hyponatremia in hospitalized patients Using Multilayer Perceptron and Multivariate Linear Regression Techniques

TL;DR

This study compares multilayer perceptron and multivariate linear regression techniques to predict hyponatremia in hospitalized patients, demonstrating that MLP significantly improves prediction accuracy over MLR.

Contribution

The paper introduces a predictive model using MLP and MLR for hyponatremia, showing MLP's superior performance in accuracy and error reduction.

Findings

MLP outperforms MLR with 23-72% higher prediction results.

MLP reduces mean squared error by 57.1% compared to MLR.

MLP achieves 27-50% higher prediction precision.

Abstract

The percentage of patients hospitalized due to hyponatremia is getting higher. Hyponatremia is the deficiency of sodium electrolyte in the human serum. This deficiency might indulge adverse effects and also associated with longer hospital stay or mortality, if it wasnt actively treated and managed. This work predicts the futuristic sodium levels of patients based on their history of health problems using multilayer perceptron and multivariate linear regression algorithm. This work analyses the patients age, information about other disease such as diabetes, pneumonia, liver-disease, malignancy, pulmonary, sepsis, SIADH, and sodium level of the patient during admission to the hospital. The results of the proposed MLP algorithm is compared with MLR algorithm based results. The MLP prediction results generates 23-72 of higher prediction results than MLR algorithm. Thus, proposed MLR algorithm has produced 57.1 of reduced mean squared error rate than the MLR results on predicting future sodium ranges of patients. Further, proposed MLR algorithm produces 27-50 of higher prediction precision rate.