2
$\begingroup$

I want to ask wich algorithm can I use to do a sequences classification , knowing that I have two classes (positive /negative), but training is done using data from one class only (positive).

Thank you

| improve this question | | | | |
$\endgroup$
2
$\begingroup$

With only one class in your training set, this sounds more like an unsupervised learning problem. Maybe look into anomaly/outlier detection algorithms rather than more traditional binary classifiers.

| improve this answer | | | | |
$\endgroup$
  • $\begingroup$ Thanks, exactly, I'm using this for anomaly detection from sequences of system calls. I thought it can be semi-supervised or unsupervised learning , but which algorithm would help? $\endgroup$ – Kahina Apr 16 '19 at 14:24
  • $\begingroup$ Ah, OK. I'm not too familiar with anomalous sequence detection, especially for nonnumerical sequences. I'd suggest editing the question to make it clearer that you're looking for something in this narrower vein, and to give some more details about your dataset; hopefully someone who is more familiar will come across the question. In the interim, a little searching came across cs.brown.edu/courses/cs227/papers/anomaly-survey-TKDE.pdf which may be helpful? Finally, as Juan said, if you have access to and actually can use the negative responses, that seems far superior. $\endgroup$ – Ben Reiniger Apr 16 '19 at 14:53
0
$\begingroup$

Is impossible to determine a model when the data you have has only one class, does not matter if they sequences, transactions, people, etc. How can your model learn the differences between categories?

The possibility you have is to run a cluster analysis on your dataset and look which variables may help you find separations between individuals, but you have to make sure that eventually you can have positives and negatives in your dataset.

| improve this answer | | | | |
$\endgroup$
  • $\begingroup$ yes, I have a dataset containing positive and negative data, but I'm using positive ones for training, and negative ones for testing $\endgroup$ – Kahina Apr 16 '19 at 14:26
  • $\begingroup$ It is incorrect. You should train your model to know the differences between positive and negative, so you should feed your model with a sample having both and test it with a different individuals which contain both too. $\endgroup$ – Juan Esteban de la Calle Apr 16 '19 at 14:30
  • $\begingroup$ Unfortunately, I don't know, the task, I'm working on, apply a behavioral approach for anomaly detection , which means to train the model whith positive behavior to detect negative behavior later. Any way, thank you for your response. $\endgroup$ – Kahina Apr 16 '19 at 14:38

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.