I need an unsupervised anomaly detection expert

AI Discussions
1

I need an expert to ask back and forth

1 Like
2

Dear @Annuse,

Feel free to share your questions or doubts, we’re here and happy to assist you.

3

I’m working on an anomaly detection system where I have multiple monitoring points (let’s call them ‘entities’) that record time-series logs over a week. Each entity has different patterns (vehicle volumes, activity cycles) and different levels of data quality (e.g., image detection confidence).

My goal is to build an Isolation Forest model for each group of similar entities.

Should I:

  • Cluster entities based on aggregated behavioral features (e.g., hourly profiles, detection success rate)?
  • Train per-entity models directly (if enough data)?

Some challenges:

  • Some entities have very noisy or unreliable data (low-quality detections)
  • Some entities have few records

What are best practices in grouping or filtering entities before building per-cluster/per-entity models?
Any tips on feature selection or dimensionality reduction in this context?

Appreciate any insights

1 Like
4

i want to know anomalies in quality

1 Like