Computers Learn To Detect Financial Abuse of the Elderly

Computer learning may provide a new avenue for creating tools to identify financial exploitation among elderly adults.

October 26, 2016

Extending on work by Drs. Shelly Jackson and Thomas Hafemister on the characteristics of elder financial exploitation,[1] NIJ funded researchers at the University of Texas Health Science Center at Houston to see if computers can “learn” how to: (1) distinguish between financial exploitation and other forms of elder abuse; and (2) differentiate between “pure” financial exploitation — when the victim of financial exploitation experiences no other forms of elder abuse and “hybrid” financial exploitation — when financial exploitation is accompanied by physical abuse or neglect.

This study demonstrated an innovative way to leverage administrative data to understand patterns of financial exploitation.

The researchers found that computer models were effective in identifying financial exploitation and its subtypes. This study may provide practitioners with ways to use existing data to identify financial exploitation among elderly adults.

Carmel Dyer, Jason Burnett, and their team used a Texas adult protective service administrative statewide dataset with 8,800 confirmed cases of elder abuse. The data were randomly split 80/20. The larger dataset was used to “train” the computer to detect patterns for financial exploitation and differentiate between pure financial exploitation and hybrid financial exploitation. The smaller dataset was used to test the computer models on accuracy in classifying the financial exploitation cases.

The computer algorithms were reliably able to predict clients who experienced financial exploitation compared with those who experienced other forms of elder abuse. Understandably, the main factors that distinguished the financial exploitation from other types of abuse were misuses of financial assets.

In distinguishing between pure financial exploitation and hybrid financial exploitation, the computer algorithms were able to make modest improvements in prediction accuracy compared to chance. The biggest factor that set the two apart was that hybrid financial exploitation cases were more likely to have an apparent injury (e.g., skin tears, bruises). Hybrid financial exploitation cases were also more likely to include clients who had overburdened caregivers, were facing a foreclosure, and were physically dependent than pure financial exploitation cases.

This study demonstrated an innovative way to leverage administrative data to understand patterns of financial exploitation.

The computer was able to learn how to distinguish financial exploitation from other types of elder abuse and further learn patterns of pure financial exploitation versus hybrid financial exploitation. The study only used the first confirmed abuse case, and there may be additional information to be learned from those who have repeat adult protective service reports. The researchers hope these data algorithms can be transformed into web-based applications so that practitioners can monitor financial exploitation in real time and quickly intervene.

About This Article

This research described in this article is based on research funded under NIJ grant 2013-IJ-CX-0050 awarded to the University of Texas Health Science Center at Houston.

The article is based on the NIJ grant final report “Exploring Elder Financial Exploitation Victimization: Identifying Unique Risk Profiles and Factors to Enhance Detection, Prevention, and Intervention” (pdf, 65 pages) by Jason Burnett, Rui Xia, Roert Suchting, and Carmel Dyer.

Cite This Article

National Institute of Justice, “Computers Learn To Detect Financial Abuse of the Elderly,” October 26, 2016, NIJ.gov: https://nij.gov/topics/crime/elder-abuse/Pages/computers-learn-to-detect-financial-abuse-of-the-elderly.aspx

Notes

[note1 ] Jackson, Shelly, Thomas L. Hafemesiter, “Financial Abuse of Elderly People vs. Other Forms of Elder Abuse: Assessing Their Dynamics, Risk Factors, and Society's Response” (pdf, 608 pages), Final Report to the National Institute of Justice, grant number 2006-WG-BX-0010, February 2011, NCJ 233613.

Date Created: October 26, 2016