Fig. 4: Rank and directionality of features predicting suicidal ideation in the best-performing classification model.
From: Neural activity during inhibitory control predicts suicidal ideation with machine learning
On the left, all features predicting the SI+/− classification in the best performing LR: Beta: IC model, ranked by average Shapley value shown with standard errors for 5-folds of cross-validation. Feature names are coded by a cognitive event within the task followed by power in a specific neural network: cue, stim, resp, or fdbk. At the right, Shapley dot plots of all ranked feature predictors show the directionality of prediction; each dot represents a single datapoint, red dots indicate larger positive feature values while blue dots indicate larger negative feature values. X axis is the Shapley values with a center point at 0. Positive Shapley values increase the model output (closer to 1 or SI+), and negative Shapley values decrease the model output (closer to 0 or SI−).
