A computerized multidimensional measurement of mental workload via handwriting analysis

Abstract

The goal of this study was to test the effect of mental workload on handwriting behavior and to identify characteristics of low versus high mental workload in handwriting. We hypothesized differences between handwriting under three different load conditions and tried to establish a profile that integrated these indicators. Fifty-six participants wrote three numerical progressions of varying difficulty on a digitizer attached to a computer so that we could evaluate their handwriting behavior. Differences were found in temporal, spatial, and angular velocity handwriting measures, but no significant differences were found for pressure measures. Using data reduction, we identified three clusters of handwriting, two of which differentiated well according to the three mental workload conditions. We concluded that handwriting behavior is affected by mental workload and that each measure provides distinct information, so that they present a comprehensive indicator of mental workload.

A New Automated Method for Evaluating Mental Workload Using Handwriting Features

“…For example, the potential of pressure-related features in detecting mental workload was not confirmed by MANOVA analyses in the study by Luria and Rosenblum [7], but their values for evaluating mental workload were clearly identified by the SVM-GA model. Pressure-related features were selected to construct the SVM-GA models for both the child (AP, SDP and S AP) and adult (AP, and S AP) data sets.…”

“…For example, there is still no general agreement on what detailed handwriting features are predictive of mental workload [6]- [8], [14]. The studies by Yu et al [8] and Lin et al [9] reported that pressure-related features were good indicators of mental workload, while Luria and Rosenblum [7] did not find significant differences in pressure-related features under three mental workload conditions. In addition, a few studies attempted to use classification models to automatically predict mental workload with some success, but they still suffered from low classification accuracy.…”

“…mean duration of pen-tip on surface and in the air) under different load conditions [7], [8], [14]. Although the empirical findings have demonstrated promise for using handwriting features to detect mental workload, there is only a limited understanding of how to use changes in handwriting behavior to indicate mental workload.…”

“…The results in Table 5 show: (1) for both the child and adult data sets, the feature subsets included four categories of features (i.e. dynamics, temporal, spatial and pressure features) and this identified the inference by Luria and Rosenblum [7] that different handwriting measures can provide distinct information on mental workload; and (2) for the child data set, two pen-orientation features (i.e. AAL and SDZA) were selected as the input of its SVM-GA model, but the feature subset for the adult data set did Table 5 The normalized features selected by the GA for each data set.…”

“…handwriting process and speech) can often be collected implicitly without extra equipment, thereby increasing their applicability in real-world environments. Recently, the feasibility of employing changes in handwriting and verbal behavior to indicate mental workload has been identified by a few empirical findings [6], [7]. In the study, we are interested in handwriting behavior because it is pervasive in current educational environments.…”