The importance of understanding and explaining the associated classification results in the utilization of artificial intelligence (AI) in many different practical applications has contributed to the trend of moving away from black-box AI towards explainable AI (XAI). In this paper, we propose the first interpretable autoencoder based on decision trees, which is designed to handle categorical data without the need to transform the data representation. Furthermore, our proposed interpretable autoencoder provides a natural explanation for experts in the application area. The experimental findings show that our proposed interpretable autoencoder is among the top-ranked anomaly detection algorithms, along with one-class SVM and Gaussian Mixture. More specifically, our proposal is on average 2\% below the best Area Under the Curve (AUC) result and 3\% over the other Average Precision scores, in comparison to One-class SVM, Isolation Forest, Local Outlier Factor, Elliptic Envelope, Gaussian Mixture Model, and eForest.

Improving fingerprint matching algorithms is an active and important research area in fingerprint recognition. Algorithms based on minutia triplets, an important matcher family, present some drawbacks that impact their accuracy, such as dependency to the order of minutiae in the feature, insensitivity to the reflection of minutiae triplets, and insensitivity to the directions of the minutiae relative to the sides of the triangle. To alleviate these drawbacks, we introduce in this paper a novel fingerprint matching algorithm, named M3gl. This algorithm contains three components: a new feature representation containing clockwise-arranged minutiae without a central minutia, a new similarity measure that shifts the triplets to find the best minutiae correspondence, and a global matching procedure that selects the alignment by maximizing the amount of global matching minutiae. To make M3gl faster, it includes some optimizations to discard non-matching minutia triplets without comparing the whole representation. In comparison with six verification algorithms, M3gl achieves the highest accuracy in the lowest matching time, using FVC2002 and FVC2004 databases.

Nowadays, the international scientific community of machine learning has an enormous campaign in favor of creating understandable models instead of black-box models. The main reason is that experts in the application area are showing reluctance due to black-box models cannot be understood by them, and consequently, their results are difficult to be explained. In unsupervised problems, where experts have not labeled objects, obtaining an explanation of the results is necessary because specialists in the application area need to understand both the applied model as well as the obtained results for finding the rationale behind each obtained clustering from a practical point of view. Hence, in this paper, we introduce a clustering based on decision trees (eUD3.5), which builds several decision trees from numerical databases. Unlike previous solutions, our proposal takes into account both separation and compactness for evaluating a feature split without decreasing time efficiency and with no empirical parameter to control the depth of the trees. We tested eUD3.5 on 40 numerical databases of UCI Machine Learning Repository, showing that our proposal builds a set of high-quality unsupervised decision trees for clustering, allowing us to obtain the best average ranking compared with other popular state-of-the-art clustering solutions. Also, from the collection of unsupervised decision trees induced by our proposal, a set of high-quality patterns are extracted for showing the main feature-value pairs describing each cluster.

Decision trees are popular as stand-alone classifiers or as base learners in ensemble classifiers. Mostly, this is due to decision trees having the advantage of being easy to explain. To improve the classification performance of decision trees, some authors have used Multivariate Decision Trees (MDTs), which allow combinations of features when splitting a node. While there is growing interest in the area, recent research in MDTs all have in common that they do not provide adequate comparison of related work: they do not consider relevant rival techniques, or they test algorithm performance in an insufficient number of databases. As a result, claims have no statistical sustain and, hence, there is a lack of general understanding of the actual capabilities of existing MDT induction algorithms, crucial to improving the state-of-the-art. In this paper, we report on an exhaustive review of MDTs. In particular, we give an overview of 37 MDT induction algorithms, out of which we have experimentally compared 19 of them in 57 databases. We provide a statistical comparison in all databases and subsets of databases according to the number of classes, number of features, number of instances, and degree of class imbalance. This allows us to identify groups of top-performing algorithms for different types of databases.