Auditing the fairness of place-based crime prediction models implemented with deep learning approaches

Jiahui Wu, Saad Mohammad Abrar, Naman Awasthi, Vanessa Frías-Martínez

Abstract: Highlights•Fairness analysis of deep-learning, crime prediction models that combine historical crime data and human mobility.•Experimental evaluation for four cities in the US: Baltimore, Minneapolis, Austin and Chicago, and eight types of crime.•Best performing models often output unfair crime predictions for groups that have historically suffered from discrimination.•Although mobility features can enhance crime prediction accuracy, they are often associated to a decrease in fairness.