|
|
|
|
In the datamining arena, 2W incorporates the SEMMA strategy coined by SAS: sampling, exploration, modification, modeling, and assessment. In the banking arena we have employed the decision tree-based models using CHAID (Chi-Square Automatic Interaction Detection) and CART (Classification and Regression Trees) that detect non-linear relationships in the data automatically to create segments. It first determines which variables are important in explaining differences according to some target variable (e.g., buying versus not buying a product). The methodology then determines how other variables behave in relation to the target, according to the level of other variables. In the final output, profiles of individuals for each of the segments and their scores are presented in graphic and language form for ease in: 1) understanding by non-technical individuals, and 2) implementation by management and staff. This is also a major breakthrough for categorical data (like gender, marital status, etc.), where 1) missing levels of categorical variables can be modeled (as opposed to deleting the entire record); and 2) past modeling efforts made it painful to determine which groups were statistically different from others. We have used this methodology further 1) in the financial services area to predict customer behavior including loan repayment/delinquency/default/ bankruptcy/successful payout/early payout); and 2) in the health care area to predict donor behavior for a non-profit foundation for target marketing. Another major benefit of this methodology is that it easily handles more than a binary response, i.e., three or more outcomes, which is more difficult to perform with traditional statistical software. We have used both SAS Enterprise Miner and SPSS AnswerTree in the financial services area to create powerful segments and profiles based on decision tree methodology. We have performed cluster analysis, unsupervised modeling, which identifies customer membership by proximity to the group center. Major and niche segments can be determined with their accompanying profiles based on the variables used in the analysis. This is very helpful to understand the customer base and to whom specific products should be marketed, based on additional information such as profitability. Other uses include determining what types of actions should be taken for specific customers. We used this methodology by applying SAS Enterprise Miner to identify customer types in the financial services area for action programs and predict their most likely type. In our work using SAS Enterprise Miner, we have also taken advantage of the powerful training/testing/validation capabilities through the splitting of the sample, i.e., data partitioning, for model creation. After models are created, their validity will be tested on the remaining data. With statistics currently, all the data are typically used for modeling, which creates upward biases in the estimation of relationships. However, the technology will allow both validity assessment and honing of models and/or estimated parameters using additional data. This is true for logistic regression, decision-trees and neural nets. We have also successfully used the replacement node to fill in missing data, a frequent problem in data analysis. Our knowledge of SAS has permitted us to program in SAS in the program code node when we were unable to perform certain data manipulations within SAS Enterprise Miner. We have also used the powerful assessment node (including lift and %capture charts) to determine what the best model and approach truly are, before scoring individuals on their propensities towards a specific behavior. Our knowledge of statistics, data mining statistical software, and base statistical software for data manipulation coupled with our extensive experience in many industries makes us your best choice for your data mining needs. Why waste time and money with less knowledgeable and less experienced SAS consultants when you can call 2W Systems Co. for your data mining application? Your North Texas SAS® Affiliate Partner
|
|
Send mail to
Info@2Wsystems.com with
questions or comments about this web site.
|
