Explaining CNN classifier using association rule mining methods on time-series

Rule-based approaches are known to be one of the most interpretable prediction models, which means that both domain experts and nondomain users can readily grasp the behavior of the model. So far, the rule mining techniques such as SBRL, Rule Regularization, and Gini Regularization have been proposed on time-series for Multilayer Perceptrons (MLPs). While MLPs perform already quite well on time-series, Convolutional Neural Networks (CNNs) outperform these algorithms in some use cases. Therefore, we adapt the existing three methods to produce an interpretable surrogate model, which imitates the behavior of the CNN classifier on time-series data. This makes the decision process more comprehensible for humans. In the experiments we evaluate three methods on the classification task trained on AIS real-world dataset and compare the performance of each method in terms of metrics such as F1-score, fidelity, and the number of rules in the surrogate list. Besides, we illustrate the impact of support, and bins (number of discrete intervals) parameters on the model's performance along with the strengths and weaknesses of each method.