Decision tree post-pruning without loss of accuracy using the SAT-PP algorithm with an empirical evaluation on clinical data

A decision tree (DT) is one of the most popular and efficient techniques in data mining. Specifically, in the clinical domain, DTs have been widely used thanks to their relatively easy explainable nature, efficient computation time, and relatively accurate predictions. However, some DT constriction algorithms may produce a large tree size structure which is difficult to understand and often leads to misclassification of data in the testing process due to poor generalization. Post pruning (PP) algorithms have been introduced to reduce the size of the tree structure with a minor (or not at all) decrease in the accuracy of classification while trying to improve the model's generalization. In this paper, we propose a new Boolean satisfiability (SAT) based PP algorithm called SAT-PP. Our algorithm reduces the tree size while preserving the accuracy of the unpruned tree. We implemented our algorithm on a medical-related classification data sets since in medical-related tasks we emphatically try to avoid decreasing the model's performance when better training is not an option. Namely, in the case of medical-related tasks, one may prefer an unpruned DT model to a pruned DT model with worse performance. Indeed, we empirically obtained that the SAT-PP algorithm produce the same accuracy and F₁ score as the DT model without PP while statistically significantly reducing the model size and as a result computation time (6.8%). In addition, we compared the proposed algorithm with other PP algorithms and found similar generalization capabilities.