Adaptive gradient boosting overview
Self-learning predictive models in Adaptive Decision Manager play a crucial part in a Next-Best-Action strategy. Adaptive models predict the propensities of all the available actions and so provide highly personalized and relevant actions to each individual customer, achieving true 1:1 customer engagement.
The adaptive gradient boosting algorithm
Adaptive gradient boosting introduces a new online learning algorithm that achieves higher predictive power. Because the predictions are more accurate, the algorithm boosts the lift in success rate and improves the customer experience, because the actions are more relevant to the customer. Adaptive gradient boosting produces more complex models, which are more difficult to explain, therefore the transparency score of this technique is lower. For more information on model transparency scores, see Configuring model transparency policy.
Advanced model settings
The adaptive gradient boosting is a self-learning algorithm, which means that it is capable of starting from a cold start, where evidence has yet to be collected, and can automatically adapt to changes in behavior over time. It builds an ensemble model, which is a combination of weak base classifiers that are decision trees. For more detailed information on how the adaptive gradient boosting works and the benefits of using it, see Adaptive Gradient Boosting - a Pega Whitepaper.
Predictor importance, also known as feature importance, allows you to gain insight into how the model is working. The predictors used by the model can be found in the model report in the Predictor importance tab. For more information on model reports, see Viewing a model report. For each predictor used by the model, an importance factor between 0 and 100 is included, ordered by decreasing importance. These factors indicate how much the score of the model is influence by each predictor. The importance factors are scaled to add up to a total of 100 for all active predictors. Predictor importance is also available at the level of an individual treatment, so you know which predictors are important to a specific treatment.
For more information about downloading a model, see Downloading the gradient booster model.
Previous topic Configuring advanced settings for an adaptive model based on gradient boosting Next topic Extracting historical responses from adaptive models