Developing Human-Centric Machine Learning Models for Temporal Data

Involving humans at every stage of developing a machine learning model is crucial for making AI systems more human-centric, both in model development and generating explanations. In this work, we developed an approach to building and iteratively improving a machine learning model with involvement of human-gained knowledge using a Spanish COVID-19 dataset as a test bed. This approach was then generalized for application to other data describing temporal phenomena, processes, or events. The proposed method utilized human insights obtained through visual analytics techniques applied to the data and the model output. By incorporating these human-gained insights into the model, performance improved and a greater understanding of the relationships between the data attributes was achieved. The insights from the COVID-19 case study were used to propose a generic workflow for developing human-centric models for temporal data. Additionally, the knowledge gained from the modeling process can potentially be used for the generation of human-centric explanations.