In total, the GP panel rated 12 statements as being important (i.e. median importance score of ≥8 and 80% consensus for the median within the panel).
We then selected the top 100, 200 and 300 features using the feature importance score to identify the most important features.
For each individual, the attribute with the highest importance score was considered the most important attribute.
For importance of change, patients were grouped as follows: "deterioration of any importance" (score: –7 to –2); "no important change" (score: –1, 0, +1); "improvement of small importance" (score: +2, +3); "improvement of moderate importance" (score: +4, +5); and "improvement of large importance" (score: +6, +7).
For a fixed number of trees, a variable with a larger importance score relative to other variables indicates that the variable is important for classification.
The random forest (RF) feature importance score was used to select a subset of features that achieved optimal accuracies.
We could have applied a cutoff to the importance score, but the scores do not have an objective scale on which to base the threshold.
We also investigated the relative importance of influencing variables and found that cohesion was the most influential variable for slope stability with an importance score of 0.310.
However, the feature importance score generated by machine learning methods is not statistically interpretable, which hampers its application in practice like medical diagnosis.
For each data set, the top 100 genes with the highest importance score were chosen, using a heuristic method aiming maximize the number of joint genes (see Methods), and then recorded and compared.
Metrics of significance include the critical value ratio, defined as the fraction of an importance score of the data item for the application of interest, and the maximum importance score computed across all the data items.
