Explainable and Local Correction of Classification Models Using Decision Trees

In practical machine learning, models are frequently updated, or corrected, to adapt new datasets. this study, we pose two challenges model correction. First, the effects of corrections end-users need be described explicitly, similar standard software where as release notes. Second, amount small so that corrected perform similarly old models. propose first correction method for classification resolves these challenges. Our idea is use an additional decision tree correct output Thanks explainability trees, describable end-users, which challenge. We resolve second challenge by incorporating when training small. Experiments on real data confirm effectiveness proposed compared existing methods.