MULTIFRACTAL DISCRIMINATION MODEL (MDM) OF HIGH-FREQUENCY PUPIL DIAMETER MEASUREMENTS FOR HUMAN-COMPUTER INTERACTION - Quantitative Medical Data Analysis Using Mathematical Tools and Statistical Techniques - page 348

Table 3. Error rate after combining the nearest neighbor classiers
rule mean median max min majority voting

Training mean 0.42 0.43 0.44 0.42 0.46

Errors std. dev. 0.01 0.01 0.02 0.02 0.01

Test mean 0.51 0.53 0.52 0.52 0.55

Errors std. dev. 0.09 0.07 0.08 0.09 0.07

Table 4. Error rate after combining the nearest neighbor classiers(with the Left Slope
feature added)

rule mean median max min majority voting

Training mean 0.407 0.414 0.417 0.401 0.432

Errors std. dev. 0.013 0.012 0.015 0.015 0.013

Test mean 0.446 0.450 0.439 0.459 0.475

Errors std. dev. 0.051 0.045 0.048 0.057 0.050

classes of users can enable developers to design improved interfaces for more
ecient and eective human-computer interactions. PRB is an informative,
yet complex, means of quantiably assessing dierences in the interaction
behaviors of users.
Using measurement of PRB during task performance is one way to study
the eects of mental workload on users. However, the inherent complexity of
PRB requires that robust and valid measures should be developed to extract
the meaningful components of the data stream in order to characterize
those changes in PRB that distinguish changes in mental workload. In this
way, the relative mental workload of users with dierent visual capabilities
can be examined. These distinctions between user needs can be used to
modify visual interfaces and interaction paradigms in order to best adapt
information technologies for users with visual impairments.
In this chapter, we study how to incorporate characteristics of the mul-
tifractal spectrum into the modeling and discrimination of the PRB high-
frequency measurement. The multifractal process was validated to be ap-
propriate in the analysis of the PRB data. The feature extraction is dis-
cussed in the context of decomposing the spectrum into describable parts.
The concepts of the spectral Mode, Broadness, and left Slope measure (the
M.B.S. summary) of a multifractal spectrum were dened. The analysis
based on the spectal Mode and Broadness measures gave distinguishable
characteristics of the PRB from the individuals with dierent visual acuity
ranges. The model-free classication method, k-nearest-neighbor classier,
is applied with the model combining technique to build a robust and accu-
rate classier.