Figure 6
The horizontal axis is labeled “S H A P value (impact on model output)” and ranges from negative 0.2 to 0.2 with an interval of 0.2. A vertical reference line at zero marks neutral impact. The vertical axis lists features in descending order of importance, including “K G C C underscore C f a”, “E 2 underscore 1”, “K G C C underscore C s a”, “Older than 35 underscore 1”, “project cost underscore Under 50 thousand dollars”, “K G C C underscore B S k”, “project end use underscore Single family or duplex dwelling”, “project cost underscore 500 thousand to 1 million”, “K G C C underscore D f b”, “K G C C underscore D f a”, “project cost underscore 250 thousand to 500 thousand”, “K G C C underscore C s b”, “project end use underscore Multi-family dwelling”, “project end use underscore Powerline, transmission line”, “project type underscore Alteration or rehabilitation”, “A 2 underscore 1”, “project end use underscore Highway, road, street”, “project end use underscore Other building”, “A 4 underscore 1”, and “A 5 underscore 1”. Each feature row contains multiple dots representing individual observations. The horizontal position of each dot corresponds to its S H A P value, showing how much that feature contributed to increasing or decreasing the prediction for that instance. The color of the dots represents the feature value, with a color bar on the right labeled “Feature value” ranging from blue (low) to pink or red (high). For the most influential features at the top, such as “K G C C underscore C f a” and “E 2 underscore 1”, points are spread more widely across the horizontal axis, indicating a stronger impact on predictions. Lower-ranked features toward the bottom show tighter clustering around zero, indicating weaker influence. The distribution of colors across each row shows how high and low feature values correspond to positive or negative contributions. Note: All numerical data values are approximated.SHAP summary plot of the RF model. Source: Authors’ own work