Predict - Best Practices

Ramya Priya Updated by Ramya Priya

Tellius offers various types of machine learning models, including regression, classification, time series regression, and clustering. Each model comes with its own unique characteristics and applications.

Best Practices

  1. Ensure the granularity of the data matches the desired outcome for predictions. If you intend to forecast monthly revenue for five different regions, you cannot use transaction-level data because the granularity of transaction data is not the same as data aggregated at the month and region levels. Aggregation functions or SQL/Python modules can help achieve the desired granularity of data.
  2. Create a separate Business View for training and testing. This allows you to explicitly control for the logic used to separate data used for model training and evaluation. It is common to use mutually exclusive date ranges for these datasets where time is an important factor in the model.
  3. Understand the business value that the model will help with. This is important for identifying the most appropriate model evaluation metric. Accuracy can be a misleading metric for model performance if the target classes are very imbalanced. AUROC is a more appropriate metric in that situation.
  4. Use the Tellius platform to handle nulls and basic feature transformations for things like one-hot encoding.
  5. Select multiple algorithms that align with your use case to identify the best-performing model. These selections are available specifically for Point-N-Click models, whereas AutoML automatically does this algorithm selection for a user.
  6. If interpretability is very critical to the business problem you are solving with a model, choosing a neural network model will make that challenging. Logistic regression or ensemble tree algorithms would be a more appropriate fit in that situation.
  7. Check the Notifications page to verify your models were trained successfully.
  8. Save the trained model object once the training process has been completed.
  9. Apply the model to the appropriate Business Views so your predictions are available for analysis by selecting the ellipses on the saved model object. The predictions are available in the Business View “Predictions” tab.
  10. Save/Move Models into Projects to organize content that can easily be shared with users or user groups.

Did we help you?

Insights (Discover) - Best Practices

Data - Best Practices