Background - I'm creating a time-series anomaly detection (TSAD) model for the wifi throughput. My customers are 2 banks, 5 retail stores, 4 universities, 6 hospitals. Currently, I have 2 options to design my solution.

  1. Separate TSAD model for each customer - I create 17 models, one for each customer. This might be most accurate but impossible to scale

  2. Industry specific TSAD model - I create 4 models. In this case, I will merge the data for 2 banks to generate 1 TSAD model. Similarly, I will merge data from 5 retail stores to generate 1 TSAD model. By repeating the process for other 2 industry verticals, I get 2 more TSAD models. This solution is better than 1. but still not great. Scalability is still a challenge because I need to maintain, update, monitor, version etc. 4 models. Moreover, the number of models will increase if model gets deployed in other industries such as Petrochemical.

Question - I would like to know is there a way for me to create one TSAD model that works for all industires. That will be truly scalable as I need to maintain, update, monitor, version one model only irrespective of the number of industries it is deployed in.

  • 1
    $\begingroup$ What exactly does this model do? What are its inputs and outputs? What specifically does it mean to "work for all industries"? Would that mean for all your customers or for every business in the world at any time or perhaps something in between? $\endgroup$
    – whuber
    Apr 19 at 18:30
  • $\begingroup$ It is an anomaly detection model that detects anomalies in the throughput of an Access Point. Because it's a time series anomaly detection, it predicts the throughput based on the past values of throughput. If the actual value deviates outside of confidence interval from the predicted value of throughput, it flags an anomaly. I have 17 customers in the 4 specified industries. I want to know how to create 1 model that would work for all 17 customers in the 4 industries I specified. Different industries have different trends. I ain't sure how to build a model that works well for all of them. $\endgroup$ Apr 19 at 23:32
  • 2
    $\begingroup$ That sounds too general to be answerable. $\endgroup$
    – whuber
    Apr 20 at 1:51
  • $\begingroup$ Can you tell me what additional information is required to help answer my question? $\endgroup$ Apr 20 at 3:50
  • 2
    $\begingroup$ It's impossible to tell. You are asking whether and how 17 models might generalize, but we don't have access to any of the details or data that would be necessary to even begin assessing them. $\endgroup$
    – whuber
    Apr 20 at 3:52


Your Answer

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

Browse other questions tagged or ask your own question.