The recent years have witnessed the rise of accurate but obscure decision systems which hide the logic of their internal decision processes to the users. The lack of explanations for the decisions of black box systems is a key ethical issue, and a limitation to the adoption of machine learning components in socially sensitive and safety-critical contexts. %Therefore, we need explanations that reveals the reasons why a predictor takes a certain decision. In this paper we focus on the problem of black box outcome explanation, i.e., explaining the reasons of the decision taken on a specific instance. We propose LORE, an agnostic method able to provide interpretable and faithful explanations. LORE first leans a local interpretable predictor on a synthetic neighborhood generated by a genetic algorithm. Then it derives from the logic of the local interpretable predictor a meaningful explanation consisting of: a decision rule, which explains the reasons of the decision; and a set of counterfactual rules, suggesting the changes in the instance's features that lead to a different outcome. Wide experiments show that LORE outperforms existing methods and baselines both in the quality of explanations and in the accuracy in mimicking the black box.

Local Rule-Based Explanations of Black Box Decision Systems.

Decision trees (DT) are a type of data classifiers. A typical classifier works in two phases. In the first, the learning phase, the classifier is built according to a preexisting data (training) set. Because decision trees are being induced from a known training set, and the labels on each example are known the first step can also be referred to as supervised learning. The second step is when the induced classifier is used for classification. Usually, prior to the first step several steps should be performed to improve the accuracy and efficiency of the classification: data cleaning, redundancy elimination, and data normalization. Classifiers are evaluated for accuracy, speed, robustness, scalability, and interpretability. DTs are widely used for exploratory knowledge discovery where comprehensible knowledge representation is preferred. The main attraction of DTs lies in the intuitive representation that is easy to understand and comprehend. Accuracy, however, is dependent on the learning data. One of the methods to improve the induction and other phases in the creation of a classifier is the use of evolutionary algorithms. They are used because the classic deterministic approach is not necessarily optimal with regard to the quality, accuracy, and complexity of the obtained classifier. In addition to the description of different evolutionary DT induction approaches, this paper also presents multiple examples of evolutionary DT applications in the medical domain. © 2012 Wiley Periodicals, Inc. © 2012 Wiley Periodicals, Inc.

Evolutionary design of decision trees for medical application

This book describes theoretical and experimental studies of instance selection to improve data mining model. Data preparation is one of the most important and time consuming phases in knowledge discovery. Preparation tasks often determine the success of data mining engagements. The importance of instance selection is the primary focus because the size of current and future databases often exceeds the amount of data which current data mining algorithms can handle properly. Instance selection thus can be used to improve scalability of data mining algorithms as well as improve the quality of the data mining results. This book presents a new optimization-based approach for instance selection that uses a genetic algorithm to select a subset of instances to produce a simpler decision tree model with acceptable accuracy. The resultant trees are easier to comprehend and interpret by the decision maker and hence more useful in practice. Numerical results are obtained for several difficult test data sets indicating that GA-based instance selection can often reduce the size of the decision tree by an order of magnitude while still maintaining good prediction accuracy.

Shuning Wu

Papers

Better Decision Tree from Intelligent Instance Selection: A new instance selection method based on Genetic Algorithm for optimizing decision trees