Titanic: The Dataset That Taught a Million People Data Science

In the early hours of April 15, 1912, the RMS Titanic sank in the North Atlantic after striking an iceberg the previous evening, killing 1,503 people. The British Board of Trade published a full passenger manifest as part of its official inquiry. The records noted class, age, sex, ticket price, cabin, and who survived. For nearly a century, this data sat in archives. Then Kaggle turned it into the most-attempted machine learning competition in history.

Kaggle launched its platform in 2010 with the goal of making data science competitions accessible. In 2012, they pinned the Titanic dataset to their front page as the "Getting Started" competition, with no prizes and no deadline. The task was simple: given the passenger data, predict who survived. Within a year, tens of thousands of people had submitted their first predictions. Within a decade, the Titanic competition had received over 15,000 teams: more entries than any other Kaggle competition in history.

The dataset is pedagogically brilliant in its imperfection. It has missing values: age is absent for 20% of passengers. It rewards feature engineering: combining sex and class into a single variable dramatically improves accuracy. It teaches overfitting: models that score perfectly on training data perform worse on test data. And it has a human truth embedded in the numbers. Women and children from first class survived at far higher rates than men from third class. "Women and children first" was policy. The data confirms it.

What Titanic gave the field is harder to measure than any accuracy score. It gave beginners a problem they already understood before they started. The historical stakes were real. The patterns were discoverable with simple tools. The intuitions were checkable: does your model agree that a first-class woman was more likely to survive than a third-class man? Probably yes. If not, something is wrong with your model. Titanic became the Hello World of data science and remains so. It is the first dataset most practitioners ever touch.