You are here:
Predicting Exposure to Consumer-Products Using Agent-Based Models Embedded with Needs-Based Artificial Intelligence and Empirically -Based Scheduling Models
Citation:
Brandon, N., P. Price, K. Isaacs, AND K. Dionisio. Predicting Exposure to Consumer-Products Using Agent-Based Models Embedded with Needs-Based Artificial Intelligence and Empirically -Based Scheduling Models. SRA 2017, Arlington, VA, December 10 - 14, 2017.
Impact/Purpose:
The modeling presented here is a significant advance in EPA's ability to characterize the use of consumer products. The work is a critical part of the Human Exposure Model of chemical exposures that happen during the use of consumer products. The research will allow EPA to better determine aggregate exposures for many chemicals.
Description:
Information on human behavior and consumer product use is important for characterizing exposures to chemicals in consumer products and in indoor environments. Traditionally, exposure-assessors have relied on time-use surveys to obtain information on exposure-related behavior. In lieu of using surveys, we create both an agent-based model (ABM) that simulates longitudinal patterns in human behavior and a scheduling model (SM) that simulates longitudinal patterns in consumer product use. By basing our ABM upon a needs-based artificial intelligence (AI) system, we create autonomous agents that mimic human decisions on residential exposure-relevant behaviors. The model predicts behavior patterns for the following actions: sleeping, eating, commuting, and working/schooling. The SM uses these behavior patterns in scheduling the use of 300+ types of consumer products during periods of idle time (i.e. when the agent is not doing the aforementioned behaviors). The ABM uses 4 different types of agents representing the following U.S. demographic groups: working and non-working adults, school-age and pre-school children. The parameters for the ABM are calibrated using survey data from the Consolidated Human Activity Database (CHAD). The SM’s predictions of daily product use are based on both seasonality of product use and also estimates of prevalence, frequency, and duration of product use derived from publicly available data on consumer habits and practices. The combined ABM/SM approach produces product use histories for individuals in the general US population that are consistent with longitudinal predictions of human behavior, reflect demographic information, and are consistent with the day of the week and season of the year. We propose that by simulating both human behavior and product use, this ABM/SM approach may allow exposure-assessors to characterize exposures from use of consumer products quicker and in ways not possible with traditional survey methods.
