Does Databricks AutoML support multi-target/multi-output classification?

[spearitchmeta]

Hi Databricks community,

I'm working on a classification problem where I need to predict 50 different target columns simultaneously using the same input features (X), commonly known as multi-target or multi-output classification.

My question: Does Databricks AutoML currently support multi-target classification, or is it limited to single-target prediction? (whether using WebUI or API)

Background:

• I have a dataset with the same feature set (X) that needs to predict multiple binary/categorical outcomes (50 Y columns)
• In traditional ML frameworks like scikit-learn, this is handled using MultiOutputClassifier or algorithms that natively support multiple outputs
• The AutoML UI and API documentation seems to only reference selecting a single "prediction target" column

Current workarounds I'm considering:

1. Running 50 separate AutoML experiments (one per target)
2. Using Databricks ML Runtime directly with scikit-learn's MultiOutputClassifier

[spearitchmeta · Accepted solution]

Long story short after lots of research: No.. Sadly this is not possible. Yet  the two mentioned options could be considered as workaround

[szymon_dybczak]

Thanks for sharing the outcome of your research @spearitchmeta. The knowledge that something is not possible is also really valuable 🙂