Functioning

The functioning of DynaMine is simple: New data mining models can be based either on the frontend by wizard or automatically created in the background. Below you find the most important functions.

Data preparation is by far the most expensive part dar in the data mining process. DynaMine takes over this task by creating an input table for each model from any data source. This process reacts flexibly to any changes in the data.

New mining models can be either created in a few steps through a wizard or created automatically. Depending on the problem it is possible to use pre-defined templates.

Because of the Design of the system it is already possible to generate very good models through the templates, a manual parameterization is possible at any time. These models will be physically applied in the respective data mining backend (SAS Enterprise Miner, Microsoft SQL Server, KNIME) and integrated into the DynaMine processes.

The scoring process applies existing models to customer data and thereby daily updated scores are created. Depending on model type these are for example buy-/churn probabilities or market basket predictions. If models are combined in an NBO approach DynaMine in addition calculates the top N product proposals.

For each model you can set a minimum performance which should not fall below this. The evaluation process of DynaMine periodically checks the predictive quality of models and automatically starts a re-training if values fall below the prescribed limit. This saves systems ressources (a retraining takes place only if it is really necessary), while a consistently high prediction quality is achieved which can not be achieved manually.

You find an overview of all mining models in the model dashboard, detailed information is presented in the reporting of DynaMine.

DynaMine Dashboard

During each training, scoring and evaluation a number of indicators is logged, so that in addition in the reporting front end the current state of the model development and its trend over time are presented.

Reports