## Context In [Hiring By Machine](https://aiethics.princeton.edu/wp-content/uploads/sites/587/2018/12/Princeton-AI-Ethics-Case-Study-5.pdf), a fictional case study from the Princeton Dialogues on AI and Ethics, developers at the company Strategeion create a machine learning system called PARiS to automatically rank job applicants based on the content of their resumes. This dataset can be used alongside the case study or as a toy dataset for exploring fairness in machine learning. ## Content The data is split into two files: - `resumes_development.csv`: 619 records used for training and validation. - `resumes_pilot.csv`: 1986 records used for the pilot phase. Each record has 222 binary features: - 218 skill features: whether or not the corresponding skill is on the applicant's resume - 4 protected features: demographic information about the applicant The columns are labeled and the list of skills can be found in `skills.txt`. The skills were selected from popular skills in [the LinkedIn directory](https://www.linkedin.com/directory/topics-a/). The four protected features are: - `Veteran`: 1 if the applicant is a veteran, 0 otherwise - `Female`: 1 if the applicant is female, 0 otherwise - `URM`: 1 if the applicant is an underrepresented minority, 0 otherwise - `Disability`: 1 if the applicant has a disability, 0 otherwise When using this dataset with the case study, the custom scikit-learn classifier `PARiSClassifier` may be used to represent the model created by the Strategeion developers. The classifier is defined in `fairness.py` and `PARiS.pickle` contains the weights for the classifier. The fictional applicant from the case study, Hara, is represented as index 1720 in the pilot dataset. ## Acknowledgements - Artificial data generated by Vinesh Kannan - Inspired by [Case Study 5: Hiring By Machine](https://aiethics.princeton.edu/case-studies/) from the Princeton Dialogues on AI and Ethics - Photo by rawpixel on Unsplash