Abstract

The maintenance of balance is one of the most important capabilities of humans. This ability is more crucial for older adults considering that over one in four elderly adults experience at least one fall (Bergen et al., 2016a) Empirical evidence showed that working memory plays role in maintaining balance. Working memory degrades by increased age. In the present study, the Dual Mechanisms of Control framework ( Braver, 2012), was utilized to evaluate the role of working memory in maintaining balance in older and younger adults. Participants were presented with visual cues indicating if the platform beneath them was likely (A) or unlikely (B) to move. On 70% of trials, cue A was followed by a forward horizontal translation of the platform. On 10% of trials, the B cue was followed by a platform movement (invalid trials). The remaining non-movement trials were either valid (B cue, 10%) or invalid (A cue, 10%) and not analyzed. Retention of goal-relevant cue information in working memory is thought to decline with aging. Kinematic signals (hip, knee, ankle joint angles) were captured by a Vicon system and characterized as (i) peak amplitude, (ii) peak latency and (iii) recovery deviation, were analyzed as a function of age group (young, older), cue type (A: likely platform movement; B: unlikely movement), and testing stage (early Block 1; late Block 6). To address the anticipated age differences in proactive control, an additional factor was considered using a neuropsychological test of working memory capacity.
Together, these four factors were included in a series of MANOVAs, using the three kinematic parameters for each of the three joint angle data. The results of multivariate ANOVAs suggested that the there was a significant effect of age and level of working memory in different age groups in some joint angles.
Furthermore, using methods of machine learning, the prediction of age group based on the kinematic characteristics was applied. Each measurement was considered as a feature and due to the excess number of features, feature selection methods (PLS, PCA, Correspondence Analysis) were applied on the dataset. The selected features by each method, composed new datasets. Three different method of Machine Learning (Decision Tree, Random Forest and Naïve Bayes) were applied to the datasets with 10-fold cross-validation. The best accuracy (0.83) was achieved by applying the Decision Tree method on a dataset selected with Partial Least Square (PLS) method. Together these results supports the Dual Mechanism of Control framework and suggests that working memory is used to maintain balance, and older adults utilized cues differently than younger adults.