Interpretable and explainable machine learning: A methods‐centric overview with concrete examples

Abstract

Interpretability and explainability are crucial for machine learning (ML) and statistical applications in medicine, economics, law, and natural sciences and form an essential principle for ML model design and development. Although interpretability and explainability have escaped a precise and universal definition, many models and techniques motivated by these properties have been developed over the last 30 years, with the focus currently shifting toward deep learning. We will consider concrete examples of state-of-the-art, including specially tailored rule-based, sparse, and additive classification models, interpretable representation learning, and methods for explaining black-box models post hoc. The discussion will emphasize the need for and relevance of interpretability and explainability, the divide between them, and the inductive biases behind the presented "zoo" of interpretable models and explanation methods.

Enhancing Generalizability in Protein–Ligand Binding Affinity Prediction with Multimodal Contrastive Learning

“…Deep learning scoring functions, while powerful, inherently have a black-box nature, making them challenging to interpret. Trusting deep learning models becomes even more complicated without an in-depth understanding of the underlying neural network mechanics . To assess whether the model learns important features aligned with physical principles, we use an approach that masks the bonds between the ligand and protein atoms.…”

“…Trusting deep learning models becomes even more complicated without an in-depth understanding of the underlying neural network mechanics. 103 To assess whether the model learns important features aligned with physical principles, we use an approach that masks the bonds between the ligand and protein atoms. This approach is designed to highlight the pivotal role of each edge concerning biological activity.…”

Classification of Kinematic Data Using Explainable Artificial Intelligence (XAI) for Smart Motion

“…Mohseni et al provide a survey and framework to evaluate XAI systems [175]. Practical XAI methods are expanded upon by Marcinkevičs and Vogt [16] and Notovich et al [176], providing application examples and a taxonomy of techniques. Domain-specific applications are explored by Preuer et al [177] in drug discovery and by Tjoa and Guan [30] in medical imaging.…”

“…They overview a myriad of interpretability methods, tailoring them for direct application within random forest models. Exploring the practical paradigms of XAI, Marcinkevičs and Vogt [16] and Notovich et al [176] present a variety of XAI methods within machine learning, while cementing these with applicationspecific examples and providing a taxonomy of different XAI techniques, respectively. The comprehensive nature of these works bridges the gap between theory and real-world applications.…”

“…As a result, explainable AI (XAI) has emerged as a crucial field, proposing solutions such as LIME [12] to improve the interpretability 1 of machine learning, potentially increasing trust in AI systems [13]. These XAI techniques not only strive for model transparency but also contribute added value to datasets, helping with tasks such as debugging [14] and localizing mislabeled samples [15], enriching the understanding of datasets and their respective domains [16], [11]. In this study, we thoroughly examine the available literature, grouping and analyzing them through the lens of our two unique observations, three purposes and four-stage data processes of XAI techniques, as follows.…”