Robust Models for Operator Workload Estimation
Abstract
When human-machine system operators are overwhelmed, judicious employment of automation can be beneficial. Ideally, a system which can accurately estimate current operator workload can make better choices when to employ automation. Supervised machine learning models can be trained to estimate workload in real time from operator physiological data. Unfortunately, estimating operator workload using trained models is limited: using a model trained in one context can yield poor estimation of workload in another. This research examines the utility of three algorithms (linear regression, regression trees, and Artificial Neural Networks) in terms of cross-application workload prediction. The study is conducted for a remotely piloted aircraft simulation under several context-switch scenarios ??? across two tasks, four task conditions, and seven human operators. Regression tree models were able to cross predict both task conditions of one task type within a reasonable level of error, and could accurately predict workload for one operator when trained on data from the other six. Six physiological input subsets were identified based on method of measurement, and were shown to produce superior cross-application models compared to models utilizing all input features in certain instances. Models utilizing only EEG features show the most potential for decreasing cross application error.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 2015
- Accession Number
- ADA615846
Entities
People
- Andrew M Smith
Organizations
- Air Force Institute of Technology