Tract details

Fill in all 13 fields. The example values come from the first row of the dataset.

The home

Average number of rooms per dwelling.

Percent of owner-occupied units built before 1940.

Land and environment

Percent of residential land zoned for lots over 25,000 sq. ft.

Percent of the town's acres used by non-retail business.

Whether the tract borders the Charles River.

Nitric oxides concentration, in parts per 10 million.

Neighborhood

Per capita crime rate by town.

Weighted distance to five Boston employment centres.

Index of accessibility to radial highways.

Full-value property tax rate per $10,000.

Pupils per teacher, by town.

Demographic variables

These two features are included only because the original dataset has them. Both are ethically problematic.

Derived from the proportion of Black residents in the town.

Why this is a problem

Percent of the population labeled "lower status".

Why this is a problem

Known issues with this dataset

The Boston Housing dataset was assembled in 1978 for a study on how air quality affects housing prices. It became a classic in machine learning tutorials, but it carries problems that this project does not fix. scikit-learn has since removed it from its library for these reasons.

B: a variable built on racial assumptions

B is calculated as 1000(Bk − 0.63)², where Bk is the proportion of Black residents in a town. It rests on the assumption that racial self-segregation raises house values. Using it as a predictor builds racial discrimination into the model.

LSTAT: people reduced to a class label

LSTAT is the percentage of the population considered "lower status", a label based on education and occupation. It reduces people to a socioeconomic category and can act as a stand-in for poverty and race.

Old and small data

The data comes from the 1970s and covers only 506 tracts around Boston. Prices are in 1970s dollars, so the predictions say nothing about today's housing market.

A model trained on this data can reproduce historical bias. This project was made for learning purposes only, to practice the full machine learning workflow from data to deployment. Its predictions must not be used for real decisions about housing, lending, or people.