I am training a classifier using SVC for a multilabel classification of texts (each text can receive a maximum of 7 labels). I am using the decision_function() method to access the least confidence instances predicted so that I'll add the instances associated with the least confidence predictions to the training set and then retrain the model. I know that the sign means on which side of the hyperplane the instance is, but I need the ones that are the closest to the hyperplane (decision boundary). An example output is below:
array([[ 0.61233108, 0.54062144, -0.66884401, -0.71035247, -0.90944953,
-0.41548433, 0.28489021],
[-0.31800646, -0.25331301, -0.40985447, -0.43378177, -1.05672207,
-0.75784478, 0.3211284 ],
[-0.12395853, 0.72147455, -0.74207976, -0.97414168, -0.9492301 ,
-0.56509169, 1.04885414]])
Which of these are the least confidence predictions? in first row the least confidence prediction would be 0.28489021 and -0.90944953 and in row 2 -1.05672207 and 0.3211284?
How to determine which are the least confidence predictions from this sample output so that I can create a function to extract them?