Intelligent Monitoring Model for Fall Risks of Hospitalized Elderly Patients.

Area of Science:

• Gerontology
• Artificial Intelligence
• Medical Informatics

Background:

• Elderly patients in hospitals face significant fall risks.
• Effective fall prevention requires early detection and real-time monitoring.
• Current monitoring systems may lack timely alarm invocation.

Purpose of the Study:

• To develop an artificial intelligence (AI) based monitoring prediction system.
• To enable early detection and real-time classification of fall risks in hospitalized elderly patients.
• To differentiate between various fall-related incidents (falls, falls with broken bones, fatal falls).

Main Methods:

• Utilized Catboost for scalable clustering and binary classification on the Snowflake platform.
• Employed a deep learning model (DNN) based on a convolutional neural network (CNN) for multi-risk classification.
• Applied adaptive sampling techniques to address dataset imbalance and overfitting.

Main Results:

• The system achieved high accuracy, with 87.4% on the SERV-112 dataset and 98.71% on the SV-S2017 dataset.
• Demonstrated higher true positive counts for all health-related risk incidences.
• Showcased real-time classification speed with reduced training time.

Conclusions:

• The proposed AI system effectively predicts and classifies fall risks in hospitalized elderly patients.
• The system offers a promising solution for real-time fall prevention and patient safety enhancement.
• The multi-risk prediction model demonstrates high performance and efficiency.