Airline Passenger Satisfaction (103K Survey Responses)

A census-scale opinion poll of flying, packaged for modeling. Airline Passenger Satisfaction (103K Survey Responses) is a labeled customer-experience panel: 129,880 surveyed passengers, each described by who they are (loyal or disloyal customer, business or personal traveler, Business, Eco or Eco Plus cabin), how far they flew, how late their flight ran, and how they scored fourteen separate touchpoints of the journey - the wifi, the booking flow, the gate, the food, the seat, the crew, the baggage belt, the cleanliness of the cabin. Every row closes with the binary verdict the whole file exists for: satisfied or neutral or dissatisfied.

That verdict column is the asset. Most customer-experience sources stop at averages; this record keeps the individual judgment attached to its causes, 129,880 times over, which is why it has become a standard classification benchmark - more than 138,000 downloads and 511 public notebooks trace back to it. It ships as a ready-made modeling split: 103,904 training rows and 25,976 test rows.

For the Passenger Airlines slice this is the only labeled experience microdata among its primary datasets - the neighbours count flights, seats and traffic, not opinions. In Datadory's catalog of 1,744 datasets across 159 viable industries the record scores 8/10 for quality with a verified field dictionary behind it. [Get a sample of this dataset](#request) and we cut it to the fields and segments you name.

Three of the shipped sample rows, exactly as the columns present themselves:

id 19556 | Female | Loyal Customer | age 52 | Business travel | Eco    | distance   160 mi
wifi 5 | schedule 4 | booking 3 | gate 4 | food 3 | online_boarding 4 | seat 3
entertainment 5 | on-board 5 | leg_room 5 | baggage 5 | checkin 2 | inflight_service 5 | cleanliness 5
dep_delay 50 min | arr_delay 44 min -> LABEL: satisfied

id 90035 | Female | Loyal Customer | age 36 | Business travel | Business | distance 2,863 mi
wifi 1 | schedule 1 | booking 3 | gate 1 | food 5 | online_boarding 4 | seat 5
entertainment 4 | on-board 4 | leg_room 4 | baggage 4 | checkin 3 | inflight_service 4 | cleanliness 5
dep_delay 0 min | arr_delay 0 min -> LABEL: satisfied

id 12360 | Male | disloyal Customer | age 20 | Business travel | Eco    | distance   192 mi
wifi 2 | schedule 0 | booking 2 | gate 4 | food 2 | online_boarding 2 | seat 2
entertainment 2 | on-board 4 | leg_room 1 | baggage 3 | checkin 2 | inflight_service 2 | cleanliness 2
dep_delay 0 min | arr_delay 0 min -> LABEL: neutral or dissatisfied

The three rows already stage the tension the full 129,880 resolve. Respondent 90035 rated the hard product lavishly - seat 5, food 5, entertainment 4 on a 2,863-mile business run - yet scored wifi 1 and schedule convenience 1, and still landed on the satisfied side of the line. Respondent 12360 scored almost nothing above 2 anywhere and dissatisfied accordingly. One row proves satisfaction survives a bad airport hour; the other shows the label tracking the average of fourteen judgments rather than any single grievance. Request a sample and the same columns come back filtered to the segment you name.

Twenty-three fields define every response, all marked verified against the published schema:

• id - respondent identifier within the extract.
• Gender - Male or Female.
• Customer Type - loyalty status: Loyal Customer or disloyal Customer.
• Age - respondent age in years.
• Type of Travel - Business travel or Personal Travel.
• Class - cabin flown: Business, Eco, or Eco Plus.
• Flight Distance - miles flown on the trip being rated.
• Fourteen service ratings on an integer 0-5 scale, where 0 means not applicable: Inflight wifi service, Departure/Arrival time convenient, Ease of Online booking, Gate location, Food and drink, Online boarding, Seat comfort, Inflight entertainment, On-board service, Leg room service, Baggage handling, Checkin service, Inflight service, Cleanliness.
• Departure Delay in Minutes and Arrival Delay in Minutes - operational context for the trip; arrival delay carries some missing values.
• satisfaction - the target label: 'satisfied' or 'neutral or dissatisfied'.

What is not there matters as much: no airline code, route, airport or flight-date column exists, which is precisely why the record works as a general experience model and cannot be used to rank a named carrier. Any carrier-level enrichment sits under additional fields on request and is confirmed against your use case before anything ships.

• Geography: not stated by the publisher - and unrecoverable from inside the files, because airline, route and airport identifiers were never included. Treat the panel as geography-blind by construction; that is the trade for a 129,880-row labeled sample.
• Temporal: a single static compilation. No per-response date is recorded, so the survey window cannot be reconstructed from the data; the record has stayed at one version since publication.
• Granularity: one row per surveyed passenger per flight experience, with all twenty-three fields populated on that row.

The consequence cuts both ways. Nothing here decays into a stale time series because there is no series to decay - the file answers "what makes a passenger satisfied" identically next year. But any question needing a when or a where needs a companion: the T-100 traffic tables or the World Bank's ICAO-sourced passengers-carried series carry the operational and geographic axes this record deliberately omits.

API, files, or your warehouse. Daily, weekly, or hourly.

You pick the channel and the cadence; the field dictionary above travels unchanged across all three. Teams benchmarking classifiers tend toward the full file load once and versioned thereafter; CX analytics groups usually scope to the attribute grid plus the segments they track; product teams embedding a satisfaction model take a scoped feed. Name the fields and filters when you [get a sample](#request) - the sample ships first either way, and changing cadence afterward is a settings conversation, not a re-integration project.

A labeled verdict on 129,880 journeys earns its keep in five specific jobs:

• Satisfaction-driver ranking - order the fourteen attributes by how strongly they move the satisfied label, so a product roadmap fixes the wifi before it repaints the lounge. The 0-5 grid makes the ranking computable rather than anecdotal.
• Classification benchmarks - train and test splits arrive pre-cut at 103,904 and 25,976 rows, so model comparisons stay like-for-like across teams and papers.
• Segment profiling - slice by Customer Type, Type of Travel and Class to watch how the same cabin scores differently for a loyal business flyer versus a disloyal personal one.
• Delay-impact analysis - join the two delay-minute columns against the label to price operational failure in satisfaction terms rather than only in compensation claims.
• Teaching and methods work - a clean, fully labeled, mid-sized tabular problem with a genuine class balance question, citable without footnote gymnastics.

Worked examples continue on the passenger airlines pages for data scientists, market researchers and competitive intelligence and product teams.

Data scientists and ML engineers get the canonical use: a large, verified, fully labeled tabular panel with the train/test split already drawn, so the pipeline starts at feature engineering instead of data assembly. Market researchers and consultants get driver analysis without a fieldwork budget - 129,880 completed interviews nobody had to recruit. Competitive intelligence and product teams get the attribute-level yardstick for experience investment cases, with the caveat that no carrier identifier exists to point it at a named rival. Journalists, academics and students get a teaching set whose provenance is documented and whose limitations are stated plainly on this page. Sales and growth teams selling CX tooling into airlines get the reference structure of a satisfaction model without licensing a vendor's black box. All delivered daily, weekly, or hourly.

Provenance note - the record circulates under the Kaggle name, uploaded in February 2020, and several near-identical copies exist on the platform under different account names. Our verification pass flagged that upstream collector and sampling frame as undocumented; cite it as a Kaggle-hosted survey unless you identify the original instrument first. The source profile covers the platform's wider catalog at Kaggle.

Quality note - Datadory scores this record 8/10 against a catalog average of 7.81 across all 1,744 datasets. Field definitions are confidence-rated verified - a bar met by 85.7% of the catalog. What holds it at 8: undocumented geography and survey period, one column with missing values, and the unresolved upstream attribution above.

Label note - the target takes exactly two values, so treat it as a classification problem, not a regression on happiness. 'Neutral or dissatisfied' is one bucket; if your use case needs the middle separated, that recoding happens on request with the schema confirmed first.

Where to go next - for the operational axis this record omits, 2015 Flight Delays and Cancellations (5.8M U.S. flights) supplies per-flight identifiers and on-time outcomes; for network-level reference data, OpenFlights airline, airport and route database maps the carriers and airports; for the macro view, the World Bank air transport passengers carried comparison sets this microdata against aggregates. The full shortlist lives at best passenger airlines datasets.