Human Workload Prediction: Lag Horizon Selection

TL;DR

Demonstrates that lag horizon selection critically shapes predictive workload forecasts derived from wearable sensors in a MATB-II supervision scenario. Compares univariate and multivariate time-series forecasts, showing univariate forecasts need longer lag horizons to reach accuracy for longer horizons, while multivariate forecasts achieve better performance with shorter lags, with diminishing returns after about 120 seconds for longer horizons. The study uses autoregressive feed-forward neural networks with cross-validation and evaluates across cognitive, visual, auditory, and overall workload components. The findings offer practical guidance for designing predictive workload systems that enable proactive robot adaptation in dynamic environments.

Abstract

Human-robot teams must be aware of human workload when operating in uncertain, dynamic environments. Prior work employed physiological response metrics from wearable sensors to estimate the current human workload; however, these estimates only enable robots to respond to under- or overload conditions reactively. Current human workload prediction approaches are limited to short prediction horizons and fail to investigate variable lag horizons' impact on predictions. This letter investigates the impact of lag horizons on both univariate and multivariate time series forecasting models for human workload prediction. A key finding is that univariate predictions required longer lag horizons of 240 seconds (s), whereas multivariate workload predictions sufficed with shorter lag horizons with diminishing returns around 120s.

Figures (1)

• Figure 1: The NASA MATB-II Tasks.