System and method for managing routing of customer calls to agents
摘要
說明書
Table 1 shows the top 15 features from an l1 buy-don't-lapse model. The most important features are identified by the highest absolute value of the importance coefficient. The most important feature of this target is the expectant_parent_nominal variable, where a 0 corresponds to not expectant. Positive and negative signs of the importance coefficient indicate whether an increases, or a decrease, of the feature increases likelihood of the target. This data indicates that non-expectant parents are less likely to buy and less likely to lapse.
In an embodiment, in building the predictive model 410, the call center evaluates performance of prospective models, such as test models, for efficacy in predicting buying behavior and/or lapse behavior. In an embodiment, prospective models are tested for the area under the curve (AUC) of a receiver-operator curve (ROC).
