Picking an optimum threshold from ROC curve

Default threshold is 0.5 for any probability based classification model. Though our prime interest in using a machine learning classification model is to jump to 0 and 1 directly, the process of how we do it from the predicted probabilities massively impact the model performance. Needless to say that the model with highest AUC ROC is best, there’s one more step after selecting the best model, that is selecting an optimum threshold. 0.5 default one is especially not best suited for class imbalance.
J-index is one of the most popular metrics used for finding optimum threshold (threshold for which J-index is maximum). J-index is defined as sensitivity + specificity -1, it boils down to TPR - FPR.