Figure 5From: Estimating tie strength in social networks using temporal communication dataResults of binary overlap prediction using single features, where the x-axis represents features, and y-axes depict: (top) Directionality of feature association, where green upwards arrows represent a positive association between the feature and the averaged overlap, red downwards arrows represent a negative association, and both green and red arrows represents non-linear association. (middle) Feature comparison with the baseline communication intensity MCC (\(\mathrm{MCC}_{*} - \mathrm{MCC}_{w}\), where ∗ is a feature) across all overlap cutoff values. A distribution fully above the red line implies a feature completely outperforms w. (bottom) Maximum MCC for four models trained with single-feature predictors, where each variable is used to predict static overlap using RF, ABC, LG and QDA. The color represents the maximum MCC over the four ML models. Variables are ranked according to their average performance over all cutoff values. For visualization purposes, we exclude weekly profile variables with an average performance of less than MCC = 0.05Back to article page