A data model is derived from transaction data. The model is represented in a combination data structure for a tree and a hash table. The hash table provides direct access to leaves of the tree, each leaf comprises a frequency count for a particular unique basket of items detected in the transaction data. Mining the combination data structure does not require recursive traversal of the tree. Moreover, derivation is performed with just two passes on the transaction data, during each pass multiple concurrent reducer tasks handle a unique portion of the transaction data providing parallel processing during creation and derivation which improves the processor elapsed time to complete the combination data structure. Furthermore, updates to the data structure are incremental without requiring any additional passes on the original transaction data and without requiring full traversal of the tree. Output from the mining is provided as input to predictor services.