Characteristic rules have been advocated for their ability to improve interpretability over discriminative rules within the area of rule learning. However, the former type of rule has not yet been used by techniques for explaining predictions. A novel explanation technique, called CEGA (Characteristic Explanatory General Association rules), is proposed, which employs association rule mining to aggregate multiple explanations generated by any standard local explanation technique into a set of characteristic rules. An empirical investigation is presented, in which CEGA is compared to two state-of-the-art methods, Anchors and GLocalX, for producing local and aggregated explanations in the form of discriminative rules. The results suggest that the proposed approach provides a better trade-off between fidelity and complexity compared to the two state-of-the-art approaches; CEGA and Anchors significantly outperform GLocalX with respect to fidelity, while CEGA and GLocalX significantly outperform Anchors with respect to the number of generated rules. The effect of changing the format of the explanations of CEGA to discriminative rules and using LIME and SHAP as local explanation techniques instead of Anchors are also investigated. The results show that the characteristic explanatory rules still compete favorably with rules in the standard discriminative format. The results also indicate that using CEGA in combination with either SHAP or Anchors consistently leads to a higher fidelity compared to using LIME as the local explanation technique.

Rules output by explainable machine learning techniques naturally come with a degree of uncertainty, as the complex functionality of the underlying black-box model often can be difficult to approximate by a single, interpretable rule. However, the uncertainty of these approximations is not properly quantified by current explanatory techniques. The use of Venn prediction is here proposed and investigated as a means to quantify the uncertainty of the explanations and thereby also allow for competing explanation techniques to be evaluated with respect to their relative uncertainty. A number of metrics of rule explanation quality based on uncertainty are proposed and discussed, including metrics that capture the tendency of the explanations to predict the correct outcome of a black-box model on new instances, how informative (tight) the produced intervals are, and how certain a rule is when predicting one class. An empirical investigation is presented, in which explanations produced by the state-of-the-art technique Anchors are compared to explanatory rules obtained from association rule mining. The results suggest that the association rule mining approach may provide explanations with less uncertainty towards the correct label, as predicted by the black-box model, compared to Anchors. The results also show that the explanatory rules obtained through association rule mining result in tighter intervals and are closer to either one or zero compared to Anchors, i.e., they are more certain towards a specific class label.

Score-based explainable machine-learning techniques are often used to understand the logic behind black-box models. However, such explanation techniques are often computationally expensive, which limits their application in time-critical contexts. Therefore, we propose and investigate the use of computationally less costly regression models for approximating the output of score-based explanation techniques, such as SHAP. Moreover, validity guarantees for the approximated values are provided by the employed inductive conformal prediction framework. We propose several non-conformity measures designed to take the difficulty of approximating the explanations into account while keeping the computational cost low. We present results from a large-scale empirical investigation, in which the approximate explanations generated by our proposed models are evaluated with respect to efficiency (interval size). The results indicate that the proposed method can significantly improve execution time compared to the fast version of SHAP, TreeSHAP. The results also suggest that the proposed method can produce tight intervals, while providing validity guarantees. Moreover, the proposed approach allows for comparing explanations of different approximation methods and selecting a method based on how informative (tight) are the predicted intervals.

Thanks to their theoretically proven properties, Shapley values have received a lot of attention as a means to explain predictions within the area of explainable machine learning. However, the computation of Shapley values is time-consuming and computationally expensive, in particular for datasets with high dimensionality, often rendering them impractical for generating timely explanations. Methods to approximate Shapley values, e.g., FastSHAP, offer a solution with adequate computational cost. However, such approximations come with a degree of uncertainty. Therefore, we propose a method to measure the fidelity of Shapley value approximations and use the conformal prediction framework to provide validity guarantees for the whole explanation in contrast to an earlier approach that offered validity guarantees on a per-feature importance basis, disregarding the relative importance of the remaining feature scores within the same explanation. We propose a set of difficulty estimation functions devised to consider the difficulty of explanation approximations. We provide a large-scale empirical investigation where the proposed difficulty estimators are evaluated with respect to their efficiency (interval size) in measuring the similarity to the ground truth Shapley values. The results suggest that the proposed approach can provide predictions coupled with informative validity guarantees (tight intervals), allowing the user to trust/reject the provided explanations based on their similarity to the ground truth values.

Shapley values have multiple desired and theoretically proven properties for explaining black-box model predictions. However, the exact computation of Shapley values can be computationally very expensive, precluding their use when timely explanations are required. FastSHAP is an approach for fast approximation of Shapley values using a trained neural network (the explainer). A novel approach, called FF-SHAP, is proposed, which incorporates three modifications to FastSHAP: i) the explainer is trained on ground-truth explanations rather than a weighted least squares characterization of the Shapley values, ii) cosine similarity is used as a loss function instead of mean-squared error, and iii) the actual prediction of the underlying model is given as input to the explainer. An empirical investigation is presented showing that FF-SHAP significantly outperforms FastSHAP with respect to fidelity, measured using both cosine similarity and Spearman's rank-order correlation. The investigation further shows that FF-SHAP even outperforms FastSHAP when using substantially smaller amounts of data to train the explainer, and more importantly, FF-SHAP still maintains the performance level of FastSHAP even when trained with as little as 15% of training data.

Data in tabular format is frequently occurring in real-world applications.Graph Neural Networks (GNNs) have recently been extended to effectively handle such data, allowing feature interactions to be captured through representation learning.However, these approaches essentially produce black-box models, in the form of deep neural networks, precluding users from following the logic behind the model predictions.We propose an approach, called IGNNet (Interpretable Graph Neural Network for tabular data), which constrains the learning algorithm to produce an interpretable model, where the model shows how the predictions are exactly computed from the original input features.A large-scale empirical investigation is presented, showing that IGNNet is performing on par with state-ofthe-art machine-learning algorithms that target tabular data, including XGBoost, Random Forests, and TabNet.At the same time, the results show that the explanations obtained from IGNNet are aligned with the true Shapley values of the features without incurring any additional computational overhead.

Many machine learning algorithms for tabular data produce black-box models, which prevent users from understanding the rationale behind the model predictions. In their unconstrained form, graph neural networks fall into this category, and they have further limited abilities to handle heterogeneous data. To overcome these limitations, an approach is proposed, called IGNH (Interpretable Graph Neural Network for Heterogeneous tabular data), which handles both categorical and numerical features, while constraining the learning process to generate exact feature attributions together with the predictions. A large-scale empirical investigation is presented, showing that the feature attributions provided by IGNH align with Shapley values that are computed post hoc. Furthermore, the results show that IGNH outperforms two powerful machine learning algorithms for tabular data, Random Forests and TabNet, while reaching a similar level of performance as XGBoost.

Shapley values have several desirable properties for explaining black-box model predictions, which come with strong theoretical support. Traditionally, Shapley values are computed post-hoc, leading to additional computational cost at inference time. To overcome this, we introduce ViaSHAP, a novel approach that learns a function to compute Shapley values, from which the predictions can be derived directly by summation. We explore two learning approaches based on the universal approximation theorem and the Kolmogorov-Arnold representation theorem. Results from a large-scale empirical investigation are presented, in which the predictive performance of ViaSHAP is compared to state-of-the-art algorithms for tabular data, where the implementation using Kolmogorov-Arnold Networks showed a superior performance. It is also demonstrated that the explanations of ViaSHAP are accurate, and that the accuracy is controllable through the hyperparameters