Bias in Facial Classification ML Models
An exploratory and statistical analysis on the biases prevalent in facial recognition machine learning models.
My team, consisting of six members, performed a litany of analyses to test if there was bias in facial recognition machine learning models. We tested the DeepFace and FairFace algorithms against a large and diverse dataset. Performing both the classic performance measurements such as accuracy and f1-score, and categorical hypothesis testing (proportionality testing), we were able to find some instances of bias. Ironically, perhaps our biggest discovery was that categorial hypoothesis testing (proportionality testing) was not a strong indicator to identify issues and errors in machine learning models.
See the links above for the complete analysis and documentation.