LoRMIkA: Local rule-based model interpretability with k-optimal associations

As we rely more and more on machine learning models for real-life\ndecision-making, being able to understand and trust the predictions becomes\never more important. Local explainer models have recently been introduced to\nexplain the predictions of complex machine learning models at the instance\nlevel. In this paper, we propose Local Rule-based Model Interpretability with\nk-optimal Associations (LoRMIkA), a novel model-agnostic approach that obtains\nk-optimal association rules from a neighbourhood of the instance to be\nexplained. Compared with other rule-based approaches in the literature, we\nargue that the most predictive rules are not necessarily the rules that provide\nthe best explanations. Consequently, the LoRMIkA framework provides a flexible\nway to obtain predictive and interesting rules. It uses an efficient search\nalgorithm guaranteed to find the k-optimal rules with respect to objectives\nsuch as confidence, lift, leverage, coverage, and support. It also provides\nmultiple rules which explain the decision and counterfactual rules, which give\nindications for potential changes to obtain different outputs for given\ninstances. We compare our approach to other state-of-the-art approaches in\nlocal model interpretability on three different datasets and achieve\ncompetitive results in terms of local accuracy and interpretability.\n