Dymatrix started in 2000 as a spinout from Computer Sciences Corporation. Focused on analytical CRM and campaign management they have done many projects and identified a number of challenges in the use of predictive analytic models. They have customers across utilities, retail, telecommunications, banking, insurance and life sciences. This work led to the development of DynaMine to automate the whole predictive analytic model lifecycle from data preparation to modeling, validation, reporting and ongoing improvement.
DynaMine is a framework for automated, adaptive model training and model management that wraps around someone else’s data mining engine and algorithms. It is designed to streamline and integrate the whole data mining process for real-time model training and scoring with any implementation of a modeling algorithm. The idea is, to use my words, to move companies from a cottage industry approach to predictive analytics to a more industrial process.
Many data mining approaches require a lot of data preprocessing, while model preparation and training require multiple manual steps. Updates are done manually (if at all) and updating models is a time consuming process. DynaMine’s framework allows for a fully automated process.
Key features:
- Pre-processing
Find new input variables and automated elimination of irrelevant variables, reduction of categorical variables etc - Adaptive Data Mining and Scoring
Automated evaluation of models and automated re-calibration - Multi-model training
Champion-challenger modeling to see which models work best - Model Management
Monitoring and reporting on model performance as well as lifecycle management and a full repository
At the core of the product is their Model Dashboard. This shows the various models as either in training (development), scoring (operation) or evaluation(monitoring).The last run and last update are also shown. The evaluation process compares results to expected and generates other reports to determine if the model is out of bounds or otherwise underperforming. Some customers re-train and re-deploy models automatically and failures can be shown in these automated processes. Others use the dashboard to see which models should be re-trained.
New models can be created using a wizard. This is based on tables defined in the DynaMine data mart and uses standard process templates defined in your modeling workbench. Parameters can be set both for the initial model build and for how and when the model should be monitored, evaluated and re-trained. The wizard creates new process flows and nodes in the data mining workbench ready for execution. The desired schedule for re-training and evaluation can also be defined including criteria for re-training based on the information available from the model training engine. Models can also be copied and used as the basis for new models.
The software supports SAS Enterprise Miner, Microsoft SQL Server Analysis Services,KNIME and IBM SPSS Modeler as underlying engines and handles predictive analytics for next best activity, affinity modeling, churn modeling, credit scoring, fraud detection, text mining etc. Integration with Zementis ADAPA allows a truly closed loop with automated model deployment to ADAPA, new results begin gathered from applications that call the ADAPA engine and automated retraining. A similar process works with generated scoring code. Both thin and fat client implementations are available.
Dymatrix is one of the vendors listed in our Decision Management Systems Platform Technologies report.
