Interactive Home Entertainment · Kaggle
Video Game Sales with Ratings
Datadory delivers interactive home entertainment data covering 16,598 commercial video games with sales above 100,000 copies - each row carrying title, release platform and year, genre, publisher, and unit sales in millions for North America, Europe, Japan, other regions and worldwide, spanning titles released from 1980 through 2016.
API, files, or your warehouse. Daily, weekly, or hourly.
- Where it covers
- Global unit sales with four-way regional split - North America, Europe, Japan, and all other regions combined - per title
- How far back
- Titles released from 1980 through 2016; the file has not been extended since October 2016, so coverage ends with mid-2010s releases
- How fine
- One row per game title per release platform, at annual release-year level; no time series within a title's life
What is the Video Game Sales with Ratings dataset?
Video Game Sales with Ratings is the Interactive Home Entertainment catalog's shared reference table: 16,598 games that each cleared 100,000 copies sold, assembled from VGChartz sales figures and published on Kaggle, where it has accumulated 802,280 downloads and 2,027 linked notebooks as of August 2026. Each row is one game-platform pairing with the title, release platform, release year, genre, publisher, and five sales figures in millions of units - North America, Europe, Japan, other regions combined, and global.
Its value is less the size than the shape: one flat CSV, one schema since 1980, no authentication wall between a question and an answer. That combination made it the default benchmark for sales forecasting, genre and platform analysis, pandas tutorials and ML coursework. The trade-off is stated plainly below - the snapshot stops in 2016 - and the fix is to treat it as a historical baseline joined against live feeds elsewhere in the same industry catalog.
Get a sample of this dataset - name the genres, platforms or decades you care about, and real rows come back before you commit to a feed.
What does a sample row look like?
Three real rows, shown exactly as they arrive:
name : Wii Sports platform : Wii year : 2006
genre : Sports publisher: Nintendo
NA 41.49M EU 29.02M JP 3.77M Other 8.46M Global 82.74M
name : Super Mario Bros. platform : NES year : 1985
genre : Platform publisher: Nintendo
NA 29.08M EU 3.58M JP 6.81M Other 0.77M Global 40.24M
name : Mario Kart Wii platform : Wii year : 2008
genre : Racing publisher: Nintendo
NA 15.85M EU 12.88M JP 3.79M Other 3.31M Global 35.82MNote what the first row implies: Global_Sales is not an independent measurement but the sum of the three named regions plus Other_Sales (41.49 + 29.02 + 3.77 + 8.46 = 82.74), so regional shares always resolve to 100 percent without re-normalizing. Note also what the third row implies - Mario Kart Wii's 3.79M Japanese units against 15.85M North American ones is the regional-skew signal the four-way split exists to surface, and it survives aggregation where averages would flatten it.
What fields does the dataset include?
Eleven verified columns per row, defined below. The load-bearing ones are Rank, a dense ordering of every row by global sales that makes percentile cuts a single predicate, and the five-column sales block, which decomposes worldwide demand into four regions whose values sum to Global_Sales by construction.
One caveat stated plainly, because the name invites the wrong expectation: despite the "with ratings" label this upload carries, the columns delivered are all sales columns. Critic and user scores are not among the eleven, and treating the field list as including them is the most common integration mistake made with this set. Score columns can be added on request from ratings-bearing derivatives in the same ecosystem.
The categorical spine - Platform, Genre, Publisher, Year - uses short controlled vocabularies (31 platforms, 12 genres), which is why cross-cuts like "Japanese-market role-playing games by publisher" resolve with one filter instead of text matching.
What does coverage look like across geography, time and granularity?
Geography: four-region decomposition per title - North America, Europe, Japan, and everything else lumped into Other_Sales. Japan is broken out separately because it behaves differently: home-console RPG lineups and handheld-first franchises produce regional skews of the kind shown in the sample rows, and a two-region model would erase them.
Time: titles released from 1980 through 2016, with the file last extended in October 2016. Thirty-six years in one schema means console generations read as a continuous series rather than stitched archives. The cutoff matters: nothing released after late 2016 appears, so the Switch generation, live-service economics and post-2017 market structure are absent by design, not omission.
Granularity: one row per game title per platform, at annual release-year level. There is no monthly or weekly series inside a title's life - the row records lifetime units by region, so use it for structural questions (which genres dominated which decades) rather than launch-window tracking.
How is the data delivered?
API, files, or your warehouse. Daily, weekly, or hourly. Pick the channel that matches the workload: an API for tools that need current rows on demand, files for batch analysis, or direct delivery into your warehouse for teams running joins against their own tables. Cadence is your call too - a static baseline suits one-off coursework, while a refreshed feed suits teams appending current-quarter releases onto the historical spine.
Start smaller than a feed. Get a sample of this dataset scoped to the genres, platforms or year ranges you care about, check that the rows answer the question, then scale the cadence.
Who uses this data, and for what?
Benchmarking a new release against history. Publishers place a title's opening numbers in context by pulling every same-genre, same-platform release since 1980 and asking where it lands in the distribution - the Rank column turns that into one predicate.
Regional portfolio strategy. The four-way sales split exposes titles that over-index in Japan or under-index in Europe, which is the input localization and publishing-rights decisions actually need.
Teaching and coursework. Eleven clean columns, no missing-key chaos and a famous top row make this the fastest way to get a class from raw file to a regression on regional preference, which is why thousands of notebooks already link to it.
Model baselines. Because the schema never changed, a model trained here reproduces identically across teams and years, making it a common yardstick when comparing forecasting approaches before they graduate to live data.
Which personas get the most value?
- Market Researchers & Consultants - sizing genre and platform markets across thirty-six years with one consistent schema.
- Competitive Intelligence & Product Teams - placing a competitor's title in the historical distribution for its genre and region mix.
- Data Scientists & ML Engineers - a clean labeled corpus for sales prediction, genre classification and reproducible benchmarking.
- Investors & Quants - franchise and publisher concentration series for gaming-sector due diligence.
- Developers & Data-Product Builders - a stable join spine of title, platform and year connecting to live player-count and chart feeds.
- Journalists, Academics & Students - citable rows behind claims about how the hit-driven games market behaved from 1980 onward.
Field dictionary
Every field below is documented against real records. The full dictionary ships with the sample.
| field | type | definition | example |
|---|---|---|---|
Rank | integer | Ranking of overall sales | 1 |
Name | string | The game's name | Wii Sports |
Platform | string | Platform of the game's release (i.e. PC, PS4, etc.) | Wii |
Year | number | Year of the game's release | 2006 |
Genre | string | Genre of the game | Sports |
Publisher | string | Publisher of the game | Nintendo |
NA_Sales | number | Sales in North America (in millions) | 41.49 |
EU_Sales | number | Sales in Europe (in millions) | 29.02 |
JP_Sales | number | Sales in Japan (in millions) | 3.77 |
Other_Sales | number | Sales in the rest of the world (in millions) | 8.46 |
Global_Sales | number | Total worldwide sales in millions of units | 82.74 |
Questions buyers ask
How many games does the Video Game Sales dataset cover?
16,598 rows, each a game-platform pairing for titles that sold more than 100,000 copies. That threshold makes it a hits-only census rather than a full catalog, which is precisely what makes it useful for studying market structure - the long tail below 100k units is excluded by construction.
Does the dataset actually include ratings?
No, despite the name. The currently served version ships eleven columns and all of them are sales or attribute columns - there are no critic or user scores among them. Ratings-bearing derivatives exist as separate uploads, so score columns can be supplied on request alongside the standard sales fields.
How far back does the sales history go?
To titles released in 1980, running through 2016 in the same schema throughout. Thirty-six years of coverage lets console generations be studied as continuous series instead of stitched-together archives, though nothing after October 2016 appears in the set.
How are sales figures expressed?
In millions of units, not revenue. Five columns carry North America, Europe, Japan, other regions combined and the global total, and the four regional figures sum to the global figure by construction, so regional share calculations need no normalization step.
Can I get a scoped sample before committing?
Yes - samples are scoped to the genres, platforms or year ranges you name, so you can check row quality against your actual question. Request one through the Get a sample of this dataset button and real rows come back, not documentation.
What can I combine this dataset with?
It pairs naturally with live feeds in the same industry: VGChartz charts and Steam concurrent-player series extend the picture past 2016, while itch.io listings add the sub-100k long tail this hits-only set excludes. Title, platform and year form a serviceable join key.
Notes on this record
- Hits-only census Every one of the 16,598 rows cleared 100,000 copies, making this a clean study population of commercial successes rather than a noisy full catalog.
- Four-region decomposition North America, Europe, Japan and other regions sum exactly to the global figure, so regional-share analysis needs no re-normalization.
- One schema since 1980 Thirty-six years of releases read as continuous series - genre lifecycles and platform transitions need no cross-archive stitching.
- Reproducible benchmark 802,280 downloads and 2,027 linked notebooks as of August 2026 mean results computed here compare directly with published work.
- Joins hold up Title plus platform plus year connects rows to VGChartz charts, Steam player counts and itch.io listings elsewhere in the same catalog.
See the rows before you pay anything.
Name this dataset and we send real records from it — scoped to the fields you asked for.