The rise of data drives the rise of Artificial Intelligence (AI). In today’s world, data is ubiquitous; however, there are gaps in the data that provide potentially sig- nificant challenges in building AI models. This work examines the development of a face database with data labeled for twelve facial attributes. This database is unique; it provides facial attributes labeled via crowdsourcing for a set of faces validated by country of origin. This effort is the first open-source dataset that observes the effect bias has in data labeling and attempts to reduce or offset the bias to provide a larger volume of accurate data for machine learning. The dataset captures images of faces from industrial and emerging countries whose facial attribute labels are validated and then separately organized by country of origin. This capstone project aims to create a process for labeling data to produce a usable country of origin image dataset for machine learning. The goal is to implement this procedure using a crowdsourcing website, The Hive AI (Hive), to accurately label thousands of images. Using this platform will improve the efficiency of this process while decreasing discrepancies in data.

Recommended Citation: Kness L., Ricanek K., Modaresnezhad M., Rachakonda L., (2023). Developing AI Data: Minimizing AI Algorithm Bias for Facial Analytics. UNCW MS CSIS Proceedings. V. 17 , N. 2 .