CAP3321C Final Project · Airports case study

Lost, Damaged, Denied

Insurance claims against U.S. airports, 2002–2009

Data Wrangling with Python · Fall 2026

Introduction · The data

Three files and nine questions

TSA claims
tsa_claims2.csv, one row per claim, columns
Global airport database
IATA code with latitude and longitude, used to place airports on the map
State outlines
Shapefile of U.S. states, read with GeoPandas
Q1Most common claim type
Q6Share of the claim that gets paid
Q2Most common claim site
Q7Five airports with the most claims
Q3Most common type at each site
Q8Total close amount over time
Q4Typical claim amount
Q9Map of claims by airport
Q5Overall approval rate
Part 1 · Data cleaning

The data and how we cleaned it

Part 1 · First look

TODO (Nicholas): headline with the main finding of Sections 2.1 and 3.1, for example why every column arrives as text

TODO (Nicholas): three or four raw values that show the problems (an amount, a date in an unusual format, a - placeholder), each with one line saying what is wrong. Copy them from your pattern_profile() tables.

Part 1 · Missing values

TODO (Nicholas): headline with the Section 3.4 finding (which years are complete and which are not)

TODO
TODO (Nicholas): claims received in the complete years
TODO
TODO (Nicholas): claims received in the years without close amounts

Part 1 · Recover before removing

TODO (Nicholas): headline on the values recovered before dropna() (outcome labels and dates)

TODO (Nicholas): Status and Disposition (Section 3.3): what the crosstab shows and how many labels are mapped.
TODO (Nicholas): dates (Section 3.5): how many receipt years and incident dates are corrected, with one before-and-after example.

Part 1 · Cleaning pipeline

Nine small functions, chained with pipe()

Part 1 · Result

TODO (Nicholas): headline with how many claims dropna() keeps and from which years

TODO
TODO (Nicholas): complete claims kept
TODO
TODO (Nicholas): incomplete rows removed, and which years they come from
Part 2 · Section 1 questions

What the claims say

Q1 · Claim type

TODO (Rayner): the Q1 answer as a headline, a full sentence with the key number

TODO
TODO (Rayner): share of the most common type
TODO
TODO (Rayner): share of the second type

Q2 · Claim site

TODO (Rayner): the Q2 answer as a headline

TODO
TODO (Rayner): claims at the most common site
TODO
TODO (Rayner): share at the second site

Q3 · Type at each site

TODO (Rayner): the Q3 answer as a headline (does the top type depend on the site?)

TODO
TODO (Rayner): top type and share at the largest site
TODO
TODO (Rayner): top type and share at the second site

Q4 · Claim amount

TODO (Rayner): the Q4 answer as a headline (typical amount, and why not the mean)

TODO
TODO (Rayner): median claim
TODO
TODO (Rayner): mean claim, and why it is higher

Q5 · Approval rate

TODO (Rayner): the Q5 answer as a headline

TODO
TODO (Rayner): approval rate (approved or settled)
TODO
TODO (Rayner): approved in full

TODO (Rayner): one line with your definition (which claims are in the denominator)

Part 3 · Section 2 questions

Payouts, airports, trends, and the map

Q6 · Share of the claim paid

Approved means paid in full; settled usually means half

of approved claims are paid exactly 100%
median payout on a settled claim; are paid exactly half
median for settled personal-injury claims

Q7 · Airports

Five large hubs account for one in five claims

of all claims at
claims at , the most of any airport

Q8 · Close amount over time

Total payouts fell from the peak

paid on claims received in
paid on claims received in 2009
→
share of claims paid anything, vs 2008: fewer payments, not smaller ones

Q9 · Building the map

Claims meet coordinates: merge on the code, then find each airport's state

rows in the airport database sit at latitude 0, longitude 0, a placeholder. Removing them makes every U.S. code unique, so the merge is one to one.
The nearest-state join keeps coastal airports such as Key West that fall just outside a simplified state outline, where a latitude/longitude box would drop them.

Q9 · The map (seaborn + GeoPandas)

claims at airports in the continental U.S.

Q9 · Interactive

The same data, live

Full screen: map.html

Conclusion

What goes wrong, where, and what gets paid

  • 1TODO (Rayner): one line on Q1-Q3, what goes wrong and where, with the key shares.
  • 2TODO (Rayner): one line on Q4-Q5, the typical claim and how often a claim is paid.
  • 3Payment is all or half. of approved claims are paid exactly in full; settled claims receive a median of .
  • 4Volume follows the hubs; payouts fell. Five hubs hold of claims, and payouts fell after as fewer claims were paid.
  • 5TODO (Nicholas): one line on the data repair and the period the analysis covers.

Thank you. Questions?