Automatic error detection from pointing device input data
Afroza Sultana, Karyn Moffatt · Proceedings of the American Society for Information Science and Technology · 2013
Abstract Although interaction techniques have been extensively studied under controlled laboratory conditions, little is known about their capabilities during unconstrained free tasks. Understanding real world use is particularly important for older adults, as some may find computer input more challenging due to age‐related declines in motor skill. As a step towards addressing this gap, we studied the feasibility of using machine‐learning techniques. In this study, we emphasized on identifying errors from sub‐movement behavior, using pen‐based data from younger and older adults. We tested four machine‐learning algorithms: Decision Trees, Neural Networks, Naïve Bayesian Networks, and Rule Induction. Each yielded an accuracy rate around 90%. Moreover, the experiments have identified some useful metrics to classify errors among older adult users.