Automatic Error Detection from Pointing Device Input Data

摘要

Although interaction techniques have been extensivelystudied under controlled laboratory conditions, little isknown about their capabilities during unconstrained freetasks. Understanding real world use is particularlyimportant for older adults, as some may find computer inputmore challenging due to age-related declines in motor skill.As a step towards addressing this gap, we studied thefeasibility of using machine-learning techniques. In thisstudy, we emphasized on identifying errors from submovementbehavior, using pen-based data from youngerand older adults. We tested four machine-learningalgorithms: Decision Trees, Neural Networks, NaïveBayesian Networks, and Rule Induction. Each yielded anaccuracy rate around 90%. Moreover, the experiments haveidentified some useful metrics to classify errors amongolder adult users.