Geek Answers Handbook

Using LIBSVM to Predict User Authentication

I plan to use LibSVM to predict user authentication in web applications. (1) Collecting data on specific user behavior (for example, login time, IP address, country, etc.) (2) Using the collected data for SVM training (3) Use real-time data to compare and generate the result at the level of authenticity.

Can someone tell me how can I do this with LibSVM? Could Century be helpful in these issues?

+7
source share
1 answer

The three steps you mentioned are the outline of the solution. In details:

  • Make sure you get a lot of tagged data, i.e. behavior logs annotated by authentic / not authentic. (Without tagged data, you find yourself in a rather advanced field of half-occupied training or must take other decisions into account.)
  • Create several functions based on data that, in your opinion, predict reliability. Try this method and refine it until it is well prepared by some statistical standard. Use a tenfold cross-check to make sure you are not overdoing it.
  • LibSVM can derive a probability estimate along with the answer; see section 8 of his manual .
+5
source

More articles:


All Articles