e-Child Happiness flagged 14 abused children whom a rule left out

South Korea's Board of Audit and Inspection found that a rule had dropped 422 children whom the e-Child Happiness model ranked higher. Fourteen of them were later found abused.

  • Not rated a person

    The 14 children came to light through reports by neighbors and others after the rule had kept them from home visits.

  • Not rated an organization

    In its re-run the board found that the model ranked children later found abused above others, but the ministry's rule kept those children from visits.

Depth runs from 1, a convenience or time saved, to 5, a life or a livelihood preserved.

Status
Partly reported benefit.
Evidence

foundstudiedreportedclaimed

An independent audit report supports the benefit.

Compared with

The 422 lower-scoring children who were selected while the exclusion rule applied. Measured.

Without the system
No confirmed abuse case among the 422 lower-scoring children
With it
14 confirmed abuse cases among the 422 higher-scoring children whom the rule had excluded
The system’s part
Not isolated. The board compared children ranked just above and just below the cut-off, which tests the model's ranking at that margin. It did not compare home visits with and without the model. The report calls the system a prediction model that weights 41 variables into a probability, but it does not say how the weights were set or whether they were learned from data.
Not yet checked
LimitsCosts and harmsLearns from dataLater outcome

Download data Cite Suggest a correction

Benefit type
Improved Task Accuracy
System
e아동행복지원시스템 (e-Child Happiness Support System) risk prediction model
Made by
Ministry of Health and Welfare (보건복지부), with data processing by the Korea Social Security Information Service (사회보장정보원)
Used by
Ministry of Health and Welfare, with local government officials who visit the selected households
Where
South Korea (national)
Stage
In routine use
Date
, when the source was published
Kind
Aggregate result
Broad benefit
Judging the Qualities of Objects, Services, or People, Analyzing Data or Information
Collections
Scrutiny and findings
Last revised
1 November 2026

Sources

  1. Audit report, by authors with no tie to the system.

    Board of Audit and Inspection of Korea (감사원). 감사보고서 - 보호대상아동 지원실태 - [Performance audit report, support for children in need of protection]. November 2019. Chapter 3, section 2, part 2-2-4, printed pp. 65-70 and [표 44].

    당초 예측모형을 통해 추출된 아동 422명의 위험 확률값이 더 높은데도 불구하고 복지사각 발굴사업에서 해당 가구가 먼저 점검되었다는 사유로 아동행복지원사업의 점검 대상에서 제외되었고, 특히 이 중 14명은 이웃 등의 신고로 아동학대로 판정되어

Details

AI system

Workflow

The system linked 41 kinds of data, such as missed infant health checks, missed medical care and utility arrears, and weighted them into a risk probability. It ranked children by that probability, and officials visited the households of those at the top, 22,727 children in the first round of 2019.

Follow-up

Conditions for replication

Children with high predicted risk must not be dropped because another program checked their households first. Officials must visit the selected households and check for signs of abuse with a checklist.

Entry details

Sector

Child protection.

Purpose

The Ministry of Health and Welfare built the e-Child Happiness Support System to predict which children are at high risk of abuse so that local officials can visit them early.

31 more questions not yet checked

model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, AI contribution, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.

Cite this record

Peter H. Vartanian. (2026). e-Child Happiness flagged 14 abused children whom a rule left out (AIBD-1012). Artificial Intelligence Benefits Database, edition of 1 November 2026. https://petervartanian.xyz/aibd/records/AIBD-1012.html (archived at https://doi.org/10.7910/DVN/AADSII, version 7.0)

BibTeX Data Field definitions