Can I Access Feature Weights in Azure AutoML Time Series Forecasting Models?

Question

After deploying a time series forecasting model in Azure AutoML, I'm interested in examining the weights assigned to each feature column utilized by the model to generate predictions. While exploring the 'Explain model' feature, I only found the weight for the target column. Is there a way to access the weights of individual feature columns to understand their significance in the model's output?

Answer 1

Hi @Anjali Kumawat,

Thank you for reaching out to Microsoft Q&A forum!

While the weights assigned to each feature column in a time series forecasting model generated by Azure AutoML are not directly accessible, but you can access the feature importance scores for each column. Feature importance scores can provide insight into the significance of each feature in the model's output.

You can access feature importance scores in Azure AutoML by following the steps provided in the Azure documentation. The feature importance scores are calculated using the permutation feature importance method, which involves randomly shuffling the values of each feature column and measuring the impact on the model's output.

You can use the feature importance scores to identify the most significant features in the model and potentially remove less significant features to improve the model's performance.

I hope you understand! Thank you.