Decision Trees
Decision trees are used to learn from historic data and to make predictions about the future. Prediction involves establishing rules using historic data and applying these rules to new data. These rules are displayed graphically as a hierarchy.
Your customer data typically contains attributes such as gender, age, income, region, and occupation as well as information about whether a customer is a satisfied customer or not (possibly drawn from a survey). You can use such historic data to train a decision tree. You find out as a result that customers exhibiting certain attributes are generally satisfied customers while customers exhibiting other attributes tend to be dissatisfied customers. You can use rules determined in this way to assess the satisfaction of other customers in cases where this information is not available.
The data that you use to train the model can be taken from any other system, provided that the system can extract data into SAP NetWeaver BW. Similarly, you can apply the identified rules to any data that has been extracted into SAP NetWeaver BW. In SAP NetWeaver BW, you can use queries to access data with known statements and then use this data to find out statements about other data.
You can make the following settings in a model for the Decision Trees method:
You use the model fields to specify which characteristic is to be considered with which attributes (such as the characteristic Customer with the attributes Occupation, Gender, Age, and so on). Moreover, you specify for which attribute the dependency on other attributes should be determined (such as the attribute Customer Satisfaction). The system then determines which of the attributes influences the dependent attribute most and takes the most influential attribute as the basis for building the decision tree.
You can use the model parameters to specify, for example, whether training should be executed using all data or whether the windowing technique should be applied to select just a representative part of the data. Furthermore, you can enhance the quality of the tree by specifying conditions for when the system should stop building the decision tree as well as by activating relevance checks and pruning.
You can display the result graphically as a hierarchy or in the form of rules. For the graphical display, you can set filters for nodes and call up detailed statistical information for individual nodes. You can also view the specific rule corresponding to a particular node in the decision tree. You must also create an analysis process to execute the prediction.