Fig. 1
From: Accelerometer-based predictive models of fall risk in older women: a pilot study
Top-ten features for two of the feature sets used in prediction of high and low function. Average importance of each feature for model prediction was computed as mean decrease impurity (see text) and is indicated by the blue bar. Black error bars represent standard deviation of importance across all trees in the forest. a Top-ten features for a random forest model trained on features extracted from individual x, y, and z axes and cross-correlations between axes. b Top-ten features for a random forest model trained on features extracted from the individual x, y, z axes, cross-correlations between axes, and traditional measures of gait
