Comparing Machine Learning Methods to Improve Fall Risk Detection in Elderly with Osteoporosis from Balance Data.

Area of Science:

• Gerontology
• Biomedical Engineering
• Data Science

Background:

• Falls are a major cause of injury in older adults, especially those with osteoporosis.
• Identifying individuals at high risk for falls is crucial for preventive interventions.

Purpose of the Study:

• To evaluate computational methods for predicting falls in older women with osteoporosis.
• To analyze the effectiveness of balance parameters in fall prediction models.

Main Methods:

• A prospective study followed 126 community-dwelling older women with osteoporosis for 2.5 years, recording falls.
• Balance parameters were measured using posturography (eyes open and closed).
• Machine learning models (IBk/KNN, Random Forest) with oversampling and feature selection were applied to predict falls.

Main Results:

• Random Forest classifier with oversampling demonstrated good predictive performance (sensitivity >0.71, specificity >0.18, PPV >0.74, NPV >0.66).
• The feature selection for minority class (FSMC) method identified novel balance parameters.
• IBk (KNN) classifier using oversampling and all variables achieved the highest performance (sensitivity >0.81, specificity >0.19, PPV =0.97, NPV =0.66).

Conclusions:

• Machine learning, particularly Random Forest and IBk (KNN) with oversampling, can effectively predict falls in older women with osteoporosis.
• Intelligent computing methods can uncover significant balance parameters often overlooked by traditional analysis.
• These models offer potential for developing predictive clinical tests to mitigate fall-related injuries.