Mitigating urinary incontinence condition using machine learning.

Area of Science:

• Biomedical Engineering
• Health Informatics
• Machine Learning

Background:

• Urinary incontinence (UI) is a condition affecting urine control, posing social, medical, and financial challenges.
• Existing applications for UI management lack predictive capabilities for urination timing.

Purpose of the Study:

• To develop a machine learning framework for predicting urination time in individuals with UI.
• To enhance the management of UI through accurate voiding predictions.

Main Methods:

• Collected 850 data points from 51 participants, recording fluid intake and urination times.
• Utilized feature selection techniques (lasso regression, decision tree, random forest, chi-square) to identify key factors influencing urination.
• Trained an extreme gradient boosting (XGB) model to classify next urination times into four categories: <30 min, 31-60 min, 61-90 min, and >90 min.

Main Results:

• Identified nine key features impacting urination frequency.
• The XGB model achieved an accuracy of 0.70, precision of 0.73, recall of 0.70, and F1 score of 0.71 in predicting urination time classes.

Conclusions:

• This study represents a foundational step in creating a predictive tool for urination timing using machine learning.
• The developed model can empower healthcare providers and caregivers with a precise instrument for assisted voiding.
• Future applications can leverage these insights to offer advanced UI management beyond current capabilities.