Project shortcuts
The challenge was to use artificial intelligence to complete different tuples of information from customer emails in such a way that at least the mandatory fields of the message capture mask were filled in. Business rules must be taken into account and missing data must be filled in. For this purpose, we have provided a Proof of Concept, whereby shipments can be filled in completely automatically based on incomplete information.
The Implementation
the evaluation was carried out in three phases
-
Data analysis of historical consignments
We paid special attention to Association Rule Mining to identify hidden rules. For this we have chosen a suitable model: Evaluation of a MissForest, which completes data by an iterative process.
-
Selection of features
We have selected features based on data analysis, as well as training and testing of the model.
-
Plausibility test of results
We tested the results to make sure the entries match. After providing the API model, the final evaluation of the prototype was carried out by our client FMS.
Result
An incomplete commit can now be sent as JSON with a POST request to the API and is returned as JSON with all required fields completed. Two models are available for imputation, which are now being tested in practice in order to support experts in the future with technically sound suggestions.