I have some samples that move on a rail for a few minutes. During this time, some forces act on these samples.
For example, I have M samples, for each sample, I have N features that are measured L times for each sample. The dimension of my data is(M*L, N).
I want to do anomaly detection to predict the anomaly before it occurs.
I have no idea how to approach this problem. Does anyone have an idea or can give me a reference to read?
I have a data set where the samples are marked as anomalous and ok.
The data looks like this:
$\begingroup$
$\endgroup$
4
-
1$\begingroup$ What is your question? This sounds like a research project that could take from a few days to a few months of work, rather than something that could be answered as a Q&A. Do you have any specific questions? $\endgroup$– TimCommented Mar 8, 2022 at 15:15
-
$\begingroup$ I need some hits to solve this problem. $\endgroup$– HanaCommented Mar 9, 2022 at 7:12
-
$\begingroup$ Can you elaborate more on what do you mean by predict anomaly before it occurs? What occurs? $\endgroup$– waicCommented Jun 20, 2022 at 9:44
-
$\begingroup$ @Hana IIUC, you have $M$ vector time series, each time series with $L$ time stamps, and each vector in those time series is of dimension $N$. Now, what do you mean by "predict before it occurs"? Do you want to have some machine that, fed with the beginning of one of your time series, will predict whether its later behavior will be anomalous? $\endgroup$– frankCommented Jul 8, 2022 at 14:03
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
There are various algorithms for anomaly detection like LOF, Isolation Forest, SVM etc. You can easily explore them on sklearn or pycaret. Further, if you are aware of the threshold values for the data, then insert them in the data so that model can easily detect them. Also, while training the data discard the Sample id and time column, otherwise it will consider them into calculation.
