Data Mining

Data Mining explained

Start with the decision the investigation should support. Do you want to understand recurring order combinations or find unusual steps in a process? Check which data actually represents that question. A large collection offers little advantage if periods, definitions or recorded events do not align. Keep origin, missing values and relevant changes traceable.

Let the task determine the method. Clustering forms groups using selected similarities; classification assigns predefined categories. Regression estimates numerical values, while association analysis searches for co-occurrence. These are different tools. A discovered customer group is not a natural, permanent property of everyone assigned to it.

Check patterns against data that did not already determine their selection or development. For predictions, use only information that would have been available at the prediction point. Otherwise an apparently accurate forecast may simply exploit later knowledge. Correlation also differs from causation, and an unusual order is not automatically fraud.

Creative Engineering connects verified analysis with an understandable application. We take responsibility for the concept and quality. An observed order combination might suggest a useful recommendation. Test that recommendation separately rather than booking the association as revenue impact. Include data maintenance, implementation and possible decision errors in the effort.