0
$\begingroup$

Problem

From what I understand, a common method in anomaly detection consists in building a predictive model trained on non-anomalous training data, and perform anomaly detection using the error of the model when predicting on the observed data. This method requires the user to identify non-anomalous data beforehand.

  • What if it's not possible to label non-anomalous data to train the model?
  • Is there anything in literature that explain how to overcome this issue?

I have an idea, but I was wondering if someone has heard for something similar before, and could point me to the right direction (link papers/blogs, or explain existing methods).

Idea

I'd like to train a decision tree on a dataset $X$ with $N$ rows, $p$ columns, having a real valued target variable $Y$ (it's a regression problem). The dataset $X$ contains both anomalous and non-anomalous objects. The decision tree training process generates groups of objects, splitting the dataset iteratively along one dimension, at each iterations. The decision trees during prediction assigns an object to a specific leaf node. Each leaf node will have a certain distribution of values of the target variable Y. An anomalous object in this problem it's the one which doesn't perform well, more precisely an object for which Y is too low.

  • Can I use the distribution in a leaf node to perform anomaly detection?
  • Assuming that a bigger value for Y is preferred, can I say that the entities in the lowest 5th percentile of Y in a node are outliers?

Example

The decision tree assigns to the node j the objects following the rule: $2 <= x_1 < 4$ and $5 <= x_2 < 7$, where $x_1$ and $x_2$ are two columns of the dataset. If I run a prediction on the entire dataset, the values of Y in the node j have have a gaussian distribution (mean=25, std=3), and I consider the values with $Y < 20.5$ outliers. The idea is that objects with $x_1$ and $x_2$ within a range, should perform similarly, and never below a target value.

Considerations

*First of all, I can see that this problem can be generalized to other methods. However, I find decision trees easier to explain here, moreover, I find the hard clustering property of trees useful for my problem, as I also need to cluster objects together. *Second, I can see the issue of training the model with both anomalous and non-anomalous data. Though, I'm wondering whether due to some property, with a big amount of data, the leaf node's distribution move's toward the optimal (value of Y of the non-anomalous objects). But I guess this depends on the ration between anomalous vs non-anomalous objects.

Any hint in the right direction would help.

$\endgroup$
2
  • $\begingroup$ It(s unclear if you have a label for your anomalies or not. Can you be more precise about that ? $\endgroup$ – lcrmorin Nov 28 '20 at 22:08
  • $\begingroup$ No, I don't have labels for the anomalies. $\endgroup$ – giogix Nov 28 '20 at 22:34
0
$\begingroup$

I will try to refer to the points I have some opinion about:

  1. "What if it's not possible to label non-anomalous data to train the model?" In this case, you face unsupervised learning problem. there are plenty of reading material regarding this topic and plenty of approaches. here is one for example: https://towardsdatascience.com/unsupervised-learning-for-anomaly-detection-44c55a96b8c1

this article may be good as well:

https://www.researchgate.net/publication/224375498_Anomaly_Detection_by_Combining_Decision_Trees_and_Parametric_Densities

  1. if you can distinguish anomalies by their distribution, it sounds that you do know the anomaly data, am I wrong?

  2. the issue with defining anomalies lower than 5th quantile, that it means that 5 % of your data that examined by the leaf will be considered as anomaly, so you if run you algorithm frequently, there is a good chance you will see a lot of anomalies. maybe try using confidence intervals.

  3. From what I have read, decision trees are not the classic method for anomaly detection. read about fraud detection, isolation trees, using SVM and there are more ways.

$\endgroup$
3
  • $\begingroup$ About the point 2) I don't know if the values below 5th percentile are anomalies, but I can assume they are, and later perform a manual check to understand if they are really anomalies. With the time I'd build a labelled dataset and I wouldn't need this method anymore ofc. This is just an initial approach. $\endgroup$ – giogix Nov 29 '20 at 18:12
  • $\begingroup$ About point 4) those methods are good, but one property I like about decision trees is that they inherently subset the dataset in clusters (I guess the same can be said about SVM). Isolation trees, though based on decision trees, do not subset the dataset based on the target variable Y. $\endgroup$ – giogix Nov 29 '20 at 18:14
  • $\begingroup$ @giogix,1. the quantile method sounds problematic to me if run your model on some time series data. I think you will get quite large amount of anomalies. 2. do you know how frequent such anomalies occur in your system? I think different methods should be considered for very rare anomalies $\endgroup$ – Arkady Mankovsky Nov 30 '20 at 8:56

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.