Campaign Manager - Campaign Manager (Silverlight)
Modeling Functionality OverviewModeling functionality provides insight into the validation of results produced. This overview provides information on the Modeling functionality in respect to the Predict Inclusion tool. The Predict Inclusion tool is based on a logistic regression algorithm. This type of regression analysis is used for predicting the outcome of a variable with two possible outcomes, for example, 'yes' or 'no', based on one or more predictor variables. Logistic regression attempts to model the probability of a 'yes/success' outcome using a linear function of the predictors. Refer to the Predictive Analytics section of the online help for details on using the Predict Inclusion tool. The purpose of this overview is to introduce the Modeling functionality and provide some insight into the validation of results produced. The Modeling algorithm is created by Extreme Optimization (www.extremeoptimization.com), it is embedded in the Campaign Manager API and accessed via the Predict Inclusion tool. This kind of regression Modeling within Campaign Management is typically used for generating Response models. These models provide a means to predict the likelihood of response to a particular campaign by modeling responders and non-responders from a previously run 'test' campaign. The output of any Predict Inclusion model is a score between 0 and 1, representing the estimated probability of the target behavior occurring if the activity (campaign) were repeated in the future. The closer the score is to 1, the more likely the behavior will be to occur. Model quality is determined by two factors:
A model that appears accurate may not be if it has weak robustness - it could produce false positives in terms of the records with higher scores when applied to other parts of the population. Conversely, a model with low accuracy adds little business value, even if it is robust. Model accuracy is primarily driven by the number and nature of the input predictor (independent) variables being used to explain the behavior shown by the target or dependent variable. Care should be taken not to 'over-fit' the model by applying too many variables, some of which have no relationship at all to the target but coincidentally have some correlation. The model robustness is generally improved by increasing the number of observation records, as the higher the number of observations, the more representative the sample is of the overall population (for example, the Customer base). Note that functionality provided by the Predict Inclusion tool, especially the 'inline' nature of its use, does NOT imply there is no longer a need for the same rigor in the development of models. It is still necessary to evaluate the results produced by the modeling process and to refine the model inputs accordingly, to produce the strongest model. It is not recommended that customer communications be generated automatically from an inline modeling process. During testing of this tool, comparisons were made between this and another respected logistic regression tool on the market, with industry recognized sample data to validate the results as being comparable and validated from a confidence perspective. Note that in those comparisons, the model score is not expected to be the same because the treatment of the variables and the modeling algorithms in each tool vary. This means that a potential campaign recipient may receive a score of 0.60 from one tool and 0.70 from another. However, it is not the individual scores produced by each tool that are being compared, but the relative score order in each case and the resulting records that are identified by the same quantiles. For example, this could be a comparison of the top deciles identified by the scores from each tool. |
| Online & Instructor-Led Courses | Training Videos | Webinar Recordings | 
|
|
|
© Alterian. All Rights Reserved. | Privacy Policy | Legal Notice |   
|