If you are debugging the Next-Best-Action decision strategy and want to see if the result of the Arbitration is correct, you should use the  Prioritize  strategy component.

Using Prediction Studio to build Pega machine learning models on historical data, you can build two types of models:  binary models  and  adaptive models .

The mapping of the input fields of a third-party predictive model is done in the Predictive Model rule. The Predictive Model rule defines how to invoke and interpret the results of a third-party predictive model that is imported in PMML format. References: https://academy.pega.com/module/predictive-analytics/topic/using-pmml-models

The principle of parsimony states that from two models with similar performance, we choose the one with the fewest number of predictors. This is because a simpler model is easier to understand, maintain, and deploy. References: https://academy.pega.com/module/predictive-analytics/topic/evaluating-predictive-models

When building a model using Pega machine learning, the validation hold-out set is used to  train the models  and to  check for robustness of candidate models .

The Monte Carlo method is a way to create a data set that includes mock-up customer data. You can use this method to generate random values for customer properties, such as age, gender, income, etc., based on predefined probabilities and distributions. References: https://academy.pega.com/module/demonstrating-adaptive-learning-archived/topic/creating-monte-carlo-data-set

According to the Pega Academy 1 , decision strategies drive the next best action and comprise a unit of reasoning represented by decision components. You use the Proposition Data component to import actions into a strategy canvas. The sequence of the components in the canvas determines which action is selected for a customer.

The Group By component 2  is used to group a list of ranked items based on a field and retain only one element in each group.  The function available when configuring the Group By component is  Count 2 , which returns the number of elements in each group.

 The value that is output by an adaptive model is behavior, which indicates the likelihood that the customer will accept or respond to an offer. Behavior is also known as propensity or probability in decision strategies. References: https://academy.pega.com/module/predicting-customer-behavior-using-real-time-data-archived/topic/using-adaptive-models-decision

A predictive model is a mathematical function that estimates the probability of an outcome based on input data.  When building a predictive model, the use of testing and validation samples  increases the accuracy of models 1 2 3 . Testing and validation samples are subsets of data that are used to evaluate how well a model performs on new data that was not used to train the model. Testing and validation samples help prevent overfitting, which is when a model learns too much from the training data and fails to generalize to new data.

Customer properties are used to map predictors to customer data that is available in the system. They can be either scalar or aggregate properties. References: https://academy.pega.com/module/creating-and-understanding-decision-strategies-archived/topic/mapping-predictors-customer