Skip to main content

Step 3: Enhance your results

This is the final step of the anonymization process.

This phase is about analyzing the results of your anonymization run and improving them — if necessary — to reach the optimal trade-off between privacy and utility according to your use case.
It’s an iterative process: by understanding the results, identifying what needs adjustment, and refining either the dataset preparation or anonymization parameters, you can progressively fine-tune your configuration.

Understand the Results​

After the run is complete, you can explore the privacy and utility metrics to evaluate the quality of your synthetic data.

➡️ See Metrics Explained

These metrics help you verify whether the anonymization strength is sufficient while retaining the necessary statistical properties for your use case.

You can also complement the built-in metrics with additional signal-specific evaluations if you have prior knowledge about key indicators you want to preserve.
The more precisely you define your expectations, the better you can guide the anonymization process toward your goals.

Identify What Needs to Be Improved and How​

There are two main types of adjustments you might want to make:

  • Improve privacy
    For example, to ensure that no individual can be identified or singled out from the synthetic dataset.

  • Improve utility
    For instance, to better preserve distributions, correlations, or predictive value.

Your next move depends on which of these you're prioritizing.

🛠️ Adapt Data Preparation and/or Parametrization​

Improving your results can be done in two main ways:

  • Optimize your data preparation
    See Prepare your data
    This might involve feature selection, variable transformation, or better handling of missing values through processors.

  • Adjust anonymization parameters
    See Run configuration
    You can change the privacy settings, influence weights, or methodology parameters.

How to Improve Privacy​

  • Increase the value of k
  • Include as many variables as possible in the anonymization process (avoid freezing too many)
  • Reduce the influence of quasi-identifiers likely to be known by attackers

How to Improve Utility​

  • Lower the value of k to allow better data fidelity
  • Freeze key variables you want to preserve as-is
  • Use the column weight parameter to give more influence to important variables in the projection space
  • Improve data preprocessing through the use of processors

🔁 Re-run and Re-evaluate​

Once you've made adjustments, you can re-launch the process using your updated setup.
Evaluate the new results, and repeat if necessary — until you reach the balance that best suits your context.

Anonymization is not one-size-fits-all: it’s about finding the right trade-off for your specific needs.