Our work / Case study
Machine Learning Decision Support for Linking Criminal History Records, With Analysts in Control
Context
What they needed
Arrest and court records sat in separate systems, with inconsistent identities and a slow, manual review process.
What we built
Ekohe built machine learning models that score likely matches, and a secure portal where analysts review, accept or reject each proposal with full traceability. More accurate record matching, with analysts in control of every decision.
Details
Analysts working on criminal history records need to connect each arrest to its court outcome. The records come from separate systems, and the same person can appear with different spellings, missing fields or inconsistent identifiers. Reviewing these links by hand was slow, and automated matches were not trusted enough to rely on.
In this context, an automated system cannot simply decide. A false match, or a missed one, changes a person's record. The platform had to make analysts faster without moving the decision away from them, and show its reasoning clearly enough to be checked.
Analysts review each proposed match with its confidence score, then accept or reject it.
Ekohe built data pipelines linking arrest records with court disposition systems, and machine learning models that compare and score each candidate match. Analysts work in a secure web portal where they review proposed matches with their confidence scores, then accept or reject them. Every decision is recorded, so the process stays traceable.
Record matching is more accurate across inconsistent datasets, and analysts spend less time on manual review. Transparent scoring gives them more confidence in each decision, and dashboards of aggregated patterns give the wider team a view they did not have before.
The confidence score is not a verdict. It tells the analyst how strong the evidence for a match is, and where to look closely. High-accountability work stays with the people accountable for it, while the model takes on the comparison work that does not scale by hand. Access to the platform is secured, and each accepted or rejected match leaves a record that can be reviewed later.
Let's START