After the features are finalized, the classifier has to be chosen which can create a model for the bot-human classification task. We have tried using various Outlier and Novelty detection techniques earlier. Since we use time sequence data as features, an RNN based classifier can yield better results than conventional classifiers (SVM, Feed Forward Neural Network).
Description
Description
Event Timeline
Comment Actions
If we could get away with just collecting high-level aggregates (e.g. the various moments of the sequence) instead the actual sequence, that would make it much easier to ensure that we are not uniquely identifying users, and a simpler classifier could be used. Although that would give very little data to train on.