I have a series of sequences, each sequence contains multiple mouse click with the following features: [x coordinate on screen, y coordinate on screen, duration of click], and each sequence is labeled with 1 of four classes which represents the user that did the sequence of clicks. I need help determining the right model/approach for this problem. Here's what I've tried so far: 1. RNN with LSTM where I had to pad the sequences with zeros since they're not equal in length 2. Draw an image with points on the coordinates clicked, and color coding the points based on the duration of the click, generating an image for each sequence, then running the images through a CNN and classifying them depending on the points

Both of these yield ~40% accuracy on submission. I need ideas/different approaches to the problem. Thank you in advance!

  • $\begingroup$ what are the classes? maybe there is a discontinuity, some click may move the page very up or very down $\endgroup$ – Fadi Bakoura May 23 '18 at 18:32
  • $\begingroup$ @FadiBakoura The classes are the names of the users that make the sequences of clicks, for example Name1 makes 50 clicks each represented by x, y, and time, Name2 makes 34 clicks, etc. There are 4 users which represent the classes. $\endgroup$ – Andrej Naumovski May 23 '18 at 18:37
  • $\begingroup$ what if we have different websites, the behavior could be totally different according to the page layout? I'm assuming the pages are websites, rights? $\endgroup$ – Fadi Bakoura May 23 '18 at 19:28

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.