100 Scanned Claims: OCR, Expense Tables, and Missing Evidence

Read scanned claim forms with OCR, rebuild their expense tables, and check your results against the originals.

advanced

Difficulty Level

5-8 hours

Estimated Time

Topics & Skills

insuranceocrjsondata-validation

Getting Started

Open the dataset folder below, grab the files, and read through the challenge. Try to solve the problem yourself before looking at the solution!

Dataset Files

Open Dataset Folder on SharePoint

Browse the folder and download the files listed in the challenge.

Scenario & Problem Statement

The challenge


These claims have been scanned, and some details are harder to read than others. Rebuild the claims and expense tables, keeping a list of anything that needs a closer look.


Start here


Open the SharePoint folder linked above. scannedClaims.zip holds the scans; the loose copies are under scannedClaims/scannedClaims. originalClaims.zip contains the original PDFs. claims.json preserves itemized expenses; claims.csv and claims.xlsx provide flattened summaries, including expense_count and expense_categories.


Preview scanned_claim (1).pdf, both its pages, and claim_0001.pdf inside originalClaims.zip. The scan has no selectable text, while the original has a text layer. Do not count archive copies and extracted copies twice.


Try it


1. Produce claims and claim_expenses tables with a review queue for uncertain fields.

2. Detect invalid archive members before OCR and reconcile the usable document count.

3. Verify every expense sum, total claimed amount, status distribution, and reporting delay.

4. Measure field accuracy against the reference and identify fields that cannot be read reliably from the scans.


What to build


Build one row per claim, one row per expense, and an audit containing source member, page, extracted value, confidence or review decision, and any answer-assisted correction. Report results separately before and after using the answer files.


Related Datasets

ACME Million: Stream Nested Sales Data and Audit Its Metadata

Work through a million sales records in small batches, check loyalty discounts, and spot mistakes in the summary.

retailbig-datajson+1

Cat Gacha: Clean Contaminated Logs and Estimate Pull Probabilities

Clean up game pull logs, remove repeated snapshots, and estimate how often each cat appears.

jsondata-cleaningprobability+1

Transaction Network 1: Join Users, Accounts, and Money Flows

Connect users, accounts, and transactions to explore a payment network and flag activity worth a closer look.

transactionsgraph-analysiscsv+1

Finished with this dataset?

Explore more datasets to continue building your data analysis and problem-solving skills.

Browse More Datasets