PEARL: Prototype Learning via Rule Learning

Deep neural networks have demonstrated promising prediction and classification performance on many healthcare applications. However, the interpretability of those models is often lacking. In comparison, classical interpretable models such as decision rule learning do not lead to the same level of accuracy as deep neural networks and can be too complex to interpret (e.g., due to large tree depths). In this work, we present PEARL, Prototype LeArNing via Rule Learning, which iteratively constructs a decision rule list to guide a neural network to learn representative prototypes. The resulting prototype neural network provides accurate prediction, and the prediction can be easily explained by a prototype and its corresponding rules. Thanks to the prediction power of neural networks and interpretability associated with rules, PEARL demonstrates state of the art accuracy to various neural networks baselines and provides simple and interpretable decision rules to explain the prediction. Experimental results also show the resulting interpretation of PEARL is simpler than the standard decision rule list while achieving much higher accuracy.