Predicting Elder Abuse Using a Random Forest Classifier.

Area of Science:

• Gerontology
• Social Sciences
• Public Health

Background:

• Elder abuse is a significant societal issue with complex underlying causes.
• Understanding the heterogeneous factors contributing to different types of elder abuse is crucial for targeted interventions.
• Traditional methods for identifying elder abuse risk factors have limitations in predictive accuracy.

Purpose of the Study:

• To identify key factors influencing elder abuse within families.
• To analyze how these factors vary across different types of elder abuse (financial, physical, psychological, neglect).
• To apply a supervised machine learning approach for enhanced risk factor identification.

Main Methods:

• Utilized data from the China Longitudinal Aging Social Survey (n=10,703).
• Employed a random forest classifier, a supervised machine learning technique.
• Identified and ranked key influencing factors based on predictive importance.

Main Results:

• Children's economic status emerged as the most significant predictor of elder abuse.
• Other key factors include number of children, older adults' health, intergenerational relations, and children's time pressure.
• Children's economic status was most predictive of physical/psychological abuse and neglect; number of children was most predictive of financial abuse.
• The impact of home-based elderly care services showed a non-monotonous relationship with elder abuse.

Conclusions:

• Machine learning, specifically random forest classification, offers superior accuracy in predicting elder abuse compared to traditional econometric models.
• Children's socioeconomic and demographic characteristics, alongside parental health and relationship quality, are critical determinants of elder abuse.
• Findings provide valuable insights for developing effective prevention and intervention strategies for elder abuse.