Catalogue of Tools & Metrics for Trustworthy AI

These tools and metrics are designed to help AI actors develop and use trustworthy AI systems and applications that respect human rights and are fair, transparent, explainable, robust, secure and safe.

CXPlain uncovers how certain factors impact housing prices in Boston

Apr 19, 2023

CXPlain uncovers how certain factors impact housing prices in Boston

How a model determines a prediction provides important context for the prediction itself. The Causal Explanations (CXPlain) method provided this context for a housing price prediction algorithm when applied to Boston housing prices. 

The Boston housing data comes from the StatLib library, which operates out of Carnegie Mellon University. The data contains 13 different features associated with houses in various locations around the Boston suburbs in the late 1970s, along with the median house price for each location. 

The people who maintain the library “strongly discourage” the use of this dataset if it is not for the purpose of “illustrating ethical issues in data science and machine learning.” That’s because a variable in the dataset “may appear to assume that racial self-segregation influences housing prices,” according to the dataset’s documentation. CXPlain was used to promote explainability of a house price prediction model, arguably to showcase that it is important to interrogate the reasoning behind the results. 

CXPlain uncovered that likely the most important features for determining housing prices in the Boston area are accessibility to a highway that leads into Boston proper and nitric oxide concentration. However, it’s worth noting that the feature importance percentage for both of these attributes have wide margins of error.

Documentation: Use case documentation

Data: Boston hosting data

Benefits of using the tool in this use case

Peeling behind the curtain of a housing price prediction model allows for interrogation of that model and the data used to fuel the results. Sharing these results in plain language can allow for broader discourse on if the right kinds and amount of data are being collected and processed.

Shortcomings of using the tool in this use case

You will need a decent understanding of data science to set the right parameters and interpret results. In this use case, only features explicitly named in the dataset were assessed, so if those features were serving as proxies to other factors not accounted for in the dataset, this analysis does not provide that information.

Learnings or advice for using the tool in a similar context

When interpreting these results, keep in mind that the dataset used is both decades old and intended for educational purposes only.

This article does not constitute legal or other professional advice and was written by students as part of the Duke Ethical Technology Practicum.

Modify this use case

About the use case


Objective(s):


Impacted stakeholders:


Target sector(s):


Country/Territory of origin: