Inquiry Regarding Suitability for Anomaly Detection

[7C00-lee]

Hello, I hope this message finds you well. I wanted to take a moment to express my sincere gratitude for your valuable contributions to GDN.

Could you kindly provide your perspective on whether your approach would be effective in detecting anomalies in the described context? Any guidance or recommendations you can offer would be greatly appreciated.

Problem description: Extract variable groups with a certain linear relationship from the data.

For example: there are 100 variables (x1 - x100), record the variable values of all variables at 100 timestamps (t1 - t100), which is a 100*100 matrix. And have a matrix that satisfies a certain relationship and a matrix that does not satisfy this relationship. The schematic diagram is as follows:

In the figure above, data 1 satisfies the relationship x3+x98 > x1+x2 from t1 to t100, while data 2 has moments when it does not satisfy this relationship (t5 - t100).

Purpose: Given the above two types of data, we can analyze which variables have a certain relationship in data 1. However, this relationship is not completely satisfied in data 2. The specific relationship does not need to be given, only the variable group is given.

Thank you for your attention to this matter. I look forward to hearing from you.

[d-ailin]

Thanks for your interest. I am not sure about whether the model can accurately capture the relationship you provided. There could be various factors affecting the captured relationship, for example, there could exist some other random variables might also be correlated to the considered variables. I think it will be better to use rule directly if you already have some strong prior information or assumptions about these variables.

[7C00-lee]

Thanks for your interest. I am not sure about whether the model can accurately capture the relationship you provided. There could be various factors affecting the captured relationship, for example, there could exist some other random variables might also be correlated to the considered variables. I think it will be better to use rule directly if you already have some strong prior information or assumptions about these variables.

Thank you for your response. It's possible that there are other variables with anomalies present simultaneously, but our aim is to capture sets containing (x1, x2, x3, x98), even if we may capture incorrect answers like (x1, x2, x3, x98, x5, x10) and (x11, x12, x13, x8, x5, x10). Do you think adding these conditions will enable GDN to produce the results we need? Thank you for your response.