Black-Box vs. White-Box: Understanding Their Advantages and Weaknesses From a Practical Point of View

Abstract

Nowadays, in the international scientific community of machine learning, there exists an enormous discussion about the use of black-box models or explainable models; especially in practical problems. On the one hand, a part of the community defends that black-box models are more accurate than explainable models in some contexts, like image preprocessing. On the other hand, there exist another part of the community alleging that explainable models are better than black-box models because they can obtain comparable results and also they can explain these results in a language close to a human expert by using patterns. In this paper, advantages and weaknesses for each approach are shown; taking into account a stateof-the-art review for both approaches, their practical applications, trends, and future challenges. This paper shows that both approaches are suitable for solving practical problems, but experts in machine learning need to understand the input data, the problem to solve, and the best way for showing the output data before applying a machine learning model. Also, we propose some ideas for fusing both, explainable and blackbox, approaches to provide better solutions to experts in real-world domains. Additionally, we show one way to measure the effectiveness of the applied machine learning model by using expert opinions jointly with statistical methods. Throughout this paper, we show the impact of using explainable and black-box models on the security and medical applications. INDEX TERMS Black-box, white-box, explainable artificial intelligence, deep learning.

An Explainable Artificial Intelligence Model for Clustering Numerical Databases

“…Decision tree (DT) algorithms have been used in several practical applications, such as bioinformatics [35], blood pressure prediction [36], agriculture [37], and travel time prediction [38], among others. The main reasons are that DTs have shown accurate results, and they provide a model easy to be interpreted by experts in the application area [39].…”

A Review of Supervised Classification based on Contrast Patterns: Applications, Trends, and Challenges

“…Examples of machine learning and deep learning approaches include random forests, gradient boosting, convolutional neural networks, and pattern recognition. Supervised classification is one of the most popular pattern recognition approaches [32,119], which has been widely studied and applied to many domains, such as bioinformatics [13,84,203], human activity recognition [94,120,146,190], rare event forecasting [34,78,162], information retrieval [18,30,171,191], face recognition [9,134,135], fingerprint identification [79,143], Internet of Things [8,202], and more recently COVID-19 (also referred to as novel Coronavirus, 2019-nCOV and SARS-CoV-2) [138,182].…”

“…Contrast pattern-based classification continues to attract interest from the research and practitioner communities since its introduction in 1963, as evidenced by the number of supervised classifiers proposed over the past decade. However, for many practical problems, obtaining a high-classification result is insufficient, because experts should also understand the classification model [46,119,127]. In many application domains, the lack of comprehensibility in the classification model(s) can result in resistance or reluctance to use certain classifiers, and in other cases, it becomes mandatory to use an understandable model.…”

“…In many application domains, the lack of comprehensibility in the classification model(s) can result in resistance or reluctance to use certain classifiers, and in other cases, it becomes mandatory to use an understandable model. For example, when credit is denied to a customer, the U.S.'s Equal Credit Opportunity Act requires financial institutions to provide reasons for rejecting the application, and vague reasons for denial are illegal [119,133]. Similarly, if the classifier is used to provide evidence in a court of law, it would be generally required to explain how the evidence was obtained.…”

Postprocessing framework for land cover classification optimization based on iterative self-adaptive superpixel segmentation

“…However, although changing the network structure can improve classification accuracy to a certain extent, neural networks have poor interpretability and operability due to the "black box" effect, and some specific problems may appear in actual application. 9 For example, when only parts of an image are poorly classified, as occurs frequently, the neural network is difficult to adjust, and retraining is very troublesome. Therefore, we think that it is meaningful if we can develop a simple efficient and human-controlled postprocessing method to optimize interior and boundary areas of classification results obtained by neural networks.…”