We have written the improved algorithm based on IBM's Cumulate algorithm, and test it on the database generated by our data generation algorithms. The future work will be to implement the improvements we propose and test it on the real medical database and other real database like Giant Eagle database.
We are also thinking about how to build a generalized decision tree, which means a generalized item could be a node in the decision tree. This seems an interesting topic since such decision tree should be more robust to noise, thus avoid the overfitting problem or save the prune stage which had to be done in ID3 or other decision tree algorithms.