Figure 5
Random forest classifier performance for different segments of each interview. (A) ROC curves for classifiers using 5 different data profiles of FX premutation carriers and comparison group. These profiles created from a different segment of each interview, but with the same length. The first and last segments of the language samples provided the most amount of information for the classifier. The information provided in segment 3 resulted in the worst performance. (B) F1 score measures the test’s accuracy considering both precision and recall. The profile constructed from the last segment has a F1 score equal to 0.74, which indicates the best performance among the tested profiles from a particular segment of an interview.
