Wavelet analysis of variance box plot.

Functional box plots satisfy two needs; visualization of functional data, and the calculation of important box plot statistics. Data visualization illuminates key characteristics of functional sets missed by statistical tests and summary statistics. The calculation of box plot statistics for functional sets permits a novel comparison more suited to functional data. The functional box plot uses a depth method to visualize and rank smooth functional curves in terms of a mean, box, whiskers, and outliers. The functional box plot improves upon other classic functional data analysis tools such as functional principal components and discriminant analysis for outlier detection. This research adds wavelet analysis as a generating mechanism along with depth for functional box plots to visualize functional data and calculate relevant statistics. The wavelet analysis of variance box plot tool gives competitive error rates in Gaussian test cases with magnitude outliers, and outperforms the functional box plot, for Gaussian test cases with shape outliers. Further, we show wavelet analysis is well suited at approximating irregular and noisy functional data and show the enhanced capability of WANOVA box plots to classify shape outliers which follow a different pattern than other functional data for both simulated and real data instances.

Modeling training effects using a human performance taxonomy.

OBJECTIVE: The aim of this study was to characterize skill acquisition during training and skill retention as a function of training strategy, retention period, and task type in the form of a numerical model and then apply that model to make predictions of performance on an unknown task. BACKGROUND: Complex systems require efficient and effective training programs for the humans who operate them in discontinuous fashion. Although there are several constructs for learning theory, models that enable analysts to predict training outcomes are needed during the design of training programs. METHOD: This study involved 60 participants who were trained on five tasks relevant to RQ-I Predator unmanned aircraft system sensor operators by one of three strategies that represented a continuum of instructor interactivity. After training, performance data for all five tasks were collected. Participants completed the same tasks 30 or 60 days later to determine skill retention and the rate at which task proficiency was reacquired. RESULTS: Models built from tasks that isolate human performance channels adequately predicted performance on a task that combined those channels. CONCLUSION: Models that predict performance on tasks that isolate human performance channels can be used to make predictions on tasks that draw on multiple channels.This model provided a distribution of performance data that was statistically similar to actual performance data. APPLICATION: System designers trained with human performance data on a set of tasks can apply those tasks' characteristics to future tasks to make reasonably accurate performance predictions, thereby allowing the designers to make early decisions regarding training strategy to teach those tasks.