Even at the best of times, traditional ways of tuning predictive models are sub optimal. There is often a period between tunings where predictions become increasingly inaccurate – but this is only discovered when the predictions can be compared with actual outcomes. Sometimes, too, the model is tuned periodically regardless of whether it needs it, wasting money.
These traditional approaches are particularly problematic during a crisis such as the pandemic. Models can appear to break down completely when presented with the disordered data collected at these times, and it’s not obvious how they can be corrected.
Stay in Tune
Capgemini’s Data Drift is a compelling approach to address this situation. This approach allows businesses to take-charge with a continuous and intelligent monitoring mechanism. This automated process, continuously monitors and triggers alerts/notifications the moment it observes abnormality in data distributions. For instance, the current pandemic situation. The data that feeds to the model is then segregated into abnormal component for analysis and normal component for prediction purposes. A clipper program snips the data into these two components to ensure the abnormality does not seep into the model outcomes.
With our Data Drift approach:
- Models are resilient: They stay relevant even in the middle of a crisis, and also as the business environment starts to recover.
- Your business need never be caught unawares by disruptive change.
- You save time and money avoiding unnecessary model tuning.
A financial services firm has a model for predicting which credit card holders will default on payments. During the pandemic, many normally reliable customers failed to pay on time owing to loss of employment, extra outgoings, and so on. Capgemini’s approach made it possible to identify defaulters who were likely to pay given the right help, such as extended payment terms. Not only was the bank able to treat each customer appropriately, but it also gained a data set with which to train its model for the recovery period.
For more information contact us:
Thank You for the submission.
We are sorry, the form submission failed. Please try again.