Operator Functional State Modeling and
Performance Prediction
The goal of this project is to develop a performance modeling and
prediction from physiological measurements.
We propose to develop a novel approach and an efficient software utility
for accurate operator functional state assessment and predication by combining
and analyzing multi-indicator signals. The approach/utility includes advanced
signal processing techniques for data preprocessing, feature analysis and
extraction, and advanced modeling technique based on Dyanmic
Bayesian Networks(DBN) for operator functional
statement modeling and prediction. This
project includes the following tasks:
o
Specifically,
through applying various signal processing and feature extraction techniques,
we will compute a set of parameters from each type of signal. Statistical analysis such as correlation
study and ANOVA will then be performed to identify a set of features that best
correlate with the operator’s functional state. We will study not only static features but
also temporal features since the operator’s functional state evolves over time,
and temporal features may be more predictive than static features.
o
Develop the
machine learning methods to learn the proposed DBN model. The model will integrate the sensory data
from different modalities both spatially and temporally to produce a composite
score to characterize the operator’s functional state and performance. For this task, we will focus on developing the
machine learning methods that can automatically learn the model structure and
parameters based on a combination of training data and the available domain
expert knowledge.
o
Demonstrate the
validity of the DBN model in operator’s state and performance prediction
through cross-validation using real data.
We
will conduct a controlled experiment to systematically compare the output of
our system with the groundtruth functional state and
with the performance measures. Specifically, we will follow the established
experimental protocol to estimate the correlation between psychomotor vigilance
test performance and the operator state parameters generated by the proposed
system for different subjects under different functional states.
The performance of our system will be characterized
by its sensitivity, specificity, and the percentage of correct detection and
misdetection. We will also perform an ROC (Receiver Operating Characteristics)
analysis and produce an ROC curve to evaluate the detection accuracy of our
system.