kanikeabhishek/toxic_comment_classification

Toxic_Comment_Classification

Our baseline models are based on logistic regression and naive bayes and a combination of the two. We then proceed to employ deep learning techniques. Further, we create an ensemble of CNN and GRU and we find that this gives the best result of all our models. CNN gave us a ROC AUC of 0.9783 whereas GRU gave us a ROC AUC of 0.981. The combination of these two gave an ROC AUC of 0.983. We expected our best model to be a blend of these techniques as some feature probabilities are better studied by some techniques than others.