Figure 14
Both plots are titled “Apartments - R F Model” and “Apartments - N N Model,” respectively. The horizontal axis in both plots represents the “SHAP value (impact on model output),” and the vertical axis lists the features considered by the models: “Thermal Satisfaction,” “Air Quality,” “Daylight Perception,” “Thermal Sensation,” “Sound Satisfaction,” “Daylight Satisfaction,” and “Sound Perception.” The color scale on the right of each plot indicates the “Feature value,” ranging from Low (blue) to High (red). In the “R F Model” plot, the horizontal axis ranges from negative 0.4 to 0.4 in increments of 0.2 units. The most impactful features are “Thermal Satisfaction” and “Sound Perception,” with distributions centered near 0 but extending broadly to both positive and negative SHAP values, indicating a complex, non-linear influence on the prediction. “Thermal Satisfaction” has the highest feature value, followed by “Sound Satisfaction,” “Air Quality,” “Daylight Perception,” and “Sound Perception.” “Thermal Sensation” has the lowest feature value. The distribution for “Daylight Satisfaction” is very narrow near 0, showing a positive SHAP value and high feature value. In the “N N Model” plot, the horizontal axis ranges from negative 0.2 to 0.6 in increments of 0.2 units. “Thermal Satisfaction” is the most influential feature, with SHAP values concentrated in the positive range (up to 0.7). Its violin plot extends widely to the right, colored predominantly red on the left and blue on the right. “Air Quality” and “Daylight Perception” are also important features, with positively extending distributions, though less pronounced than “Thermal Satisfaction.” Most features are centered around 0 and show high feature values near this point. Note: All numerical values are approximate.Violin plots of SHAP values illustrating each feature's contribution to the predicted IE satisfaction in apartments ML models. Source: Authors' own work