I-RF: A Transparent Decision-Making System

Ensemble learning is a hybrid learning system that exhibits a high level of performance by aggregating predictions of multiple classifiers and hence used to obtain a powerful predictive performance. Random Forest (RF) is an ensemble learning technique which executes a huge number of Decision Tree (DT) based on different subset of data and feature combinations. Regardless of its high-performance, RF is black box in nature which hinders the interpretability of the predictive model. A transparent system with less decision rules makes a system efficient, user convincing and manageable to a greater extent in fields like medical, business, banking etc. The expression of the decision rules into flowchart like representation makes the system transparent, explicitly understandable and closely resemblance to human reasoning. Therefore, to overcome the disadvantages of black box nature and to make it an efficient interpretable decision-making system, this paper proposes a transparent RF named Interpretable RF (I-RF) using Significance Score by combining Accuracy and Transparency to extract the important decision rules hence making RF behaves like a white box which is transparent and comprehensible. The proposed model I-RF is compared with the performances of a simple DT and RF, Support Vector Machine (SVM) and Naïve Bayes in terms of classification accuracy, precision, recall and F1 score measures. In addition, I-RF is also compared with TRG-RF, RuleFit and RF + DHC which are also rule-based methods. The performance of the proposed I-RF is validated with 12 well known UCI datasets and Kaggle and is observed from the experimental that the proposed I-RF is more efficient interpretable decision-making system.