Integrating Human Behavior Modeling and Data Mining Techniques to Predict Human Errors in Numerical Typing

Numerical typing errors can lead to serious consequences, but various causes of human errors and the lack of contextual clues in numerical typing make their prediction difficult. Human behavior modeling can predict the general tendency in making errors, while data mining can recognize neurophysiological feedback in detecting cognitive abnormality on a trial-by-trial basis. This study suggests integrating human behavior modeling and data mining to predict human errors because it utilizes both 1) top-down inference to transform interactions between task characteristics and conditions into a general inclination of an average operator to make errors and 2) bottom-up analysis in parsing psychophysiological measurements into an individual's likelihood of making errors on a trial-by-trial basis. Real-time electroencephalograph (EEG) features collected in a numerical typing experiment and modeling features produced by an enhanced human behavior model (queuing network model human processor) were combined to improve error classification performance by a linear discriminant analysis (LDA) classifier. Integrating EEG and modeling features improved the results of LDA classification by 28.3% in keenness (d') and by 10.7% in the area under ROC curve (AUC) from that of using EEG only; it also outperformed the other three benchmarking scenarios: using behaviors only, using apparent task features, and using task features plus trial information. The AUC was significantly increased from using EEG along only if EEG + Model features were used.