— This study proposes a scalable explainable artificial intelligence (XAI) framework for multivariate time-series classification by addressing the computational limitations of SHAP-based temporal attribution methods. While approaches such as Temporal SHAP , TimeSHAP , and Dynamic WindowSHAP provide window-level interpretability, their combinatorial complexity grows rapidly with sequence length and feature dimensionality, limiting practical applicability. To overcome these limitations, we introduce a perturbation-driven dynamic window framework that estimates temporal importance by directly measuring model output deviations under structured temporal perturbations, thereby avoiding explicit Shapley-value enumeration. This approximation enables efficient temporal attribution while preserving dominant importance patterns observed in SHAP-based methods. The proposed unified framework consists of four components: (1) Dynamic WindowPerturbation for efficient window-level attribution, (2) PCA-integrated dynamic window analysis to reduce feature redundancy while reconstructing attributions in the original feature space, (3) concept-level temporal attribution based on time-series decomposition, and (4) global temporal aggregation for dataset-level interpretability. Experiments on a synthetic MIMIC-III clinical dataset and a real Swarm Behaviour dataset using GRU and LSTM models demonstrate up to 89% reduction in explanation runtime compared to SHAP-based methods and up to 9 × acceleration through PCA integration. Independent faithfulness evaluation using deletion and insertion metrics confirms that the perturbation-based approach preserves decision-relevant temporal importance structures despite relaxing Shapley axioms. Overall, the results demonstrate that dominant temporal importance patterns in high-dimensional sequential models can be recovered without exact Shapley-value computation, enabling a practical trade-off between axiomatic guarantees and computational scalability.

Explainable Artificial Intelligence (XAI) plays a crucial role in high-stakes decision-making by ensuring that machine learning models provide clear and trustworthy explanations. However, many existing interpretability methods, such as SHAP and Partial Dependence Plot (PDP), struggle to differentiate between correlation and causality. To overcome this challenge, we introduce a causality-aware surrogate modeling framework that improves the global interpretability of complex models. Our approach combines Probability of Sufficiency (PS), Probability of Necessity (PN), and Probability of Causality (PoC) with decision tree-based rule extraction to identify rules and features that have a direct causal impact on the target outcome. Experiments on the German Credit dataset reveal that certain rules exhibit strong sufficiency while showing weak necessity, highlighting key causal factors in loan approval decisions. Among these, savings and duration stand out as critical features for counterfactual reasoning. By ensuring that extracted rules capture true causal relationships rather than misleading correlations, our method enhances model transparency, trustworthiness, and counterfactual fundamentals.

ABSTRACT The field of recommendation systems is a hot topic thanks to the increasing number of available digital products and services. In connection with this topic, the research of Graph Neural Network solutions has played a significant role in recent years. Research and development of an online recommendation system that also manages the challenges of a rapidly changing environment are important from a practical point of view as well. Our aim was to develop an approach that possesses scalable inference and adaptation and uses latent features. The main contribution of this paper is the development of a candidate generation process for online collaborative filtering on implicit feedback data that can scale to large user and item bases. We proposed multiple ways how embeddings can be obtained in a fast and scalable way, namely Lookup, Inductive neighbor aggregation, Neighbor aggregation with importance scores, and GraphSAGE‐based Graph Neural Network (GraphSAGE+) method with continuous representation update for online learning. By combining these inductive and transductive methods for the embeddings, we developed a novel online Collaborative Filtering approach. We evaluated our approach on two e‐commerce datasets and found that it outperformed traditional recommendation algorithms such as Matrix Factorization.

Abstract Concept drift (and data drift) is a common phenomenon in machine learning models, where the statistical properties of the input data change over time, leading to a decrease in model performance. Detecting data drift is crucial for maintaining the accuracy and reliability of machine learning models in real-world applications. While previous data drift detector approaches can identify if a drift has occurred, these approaches cannot localize which specific features have caused the drift. Feature drift detectors solve this deficiency, but the required number of detectors is equal to the number of dimensions, which is a resource-intensive solution in high-dimensional data. In this paper, we propose a novel approach for feature drift analysis and drift detection based on a domino effect caused by the correlation of features. Our approach, the so-called Domino drift effect (DDE), is based on the empirically proven assumption that an initial reference correlation can be utilized as a proxy for detecting other drifting features. The method analyzes the correlating and drifting behavior, and by using only a subset of all features, it derives inference about the drifting of the remaining features, if co-drifting phenomena occur in the data stream. At co-drifting phenomena, the DDE method can estimate the probability of feature drift, which is particularly useful in high-dimensional datasets. To evaluate the effectiveness of our approach, we conducted experiments on four real-world datasets. The results show that our approach can effectively be used to predict feature drift in the whole dataset, and it has potential industrial applications.

Abstract The fast improvement of deep learning methods resulted in breakthroughs in image classification, object detection, and object tracking. Autonomous driving and traffic monitoring systems, especially the on-premise installed fixed position multi-camera configurations, benefit greatly from recent advances. In this paper, we propose a Multi-Camera Multi-Target (MCMT) vehicle tracking system using a constrained hierarchical clustering solution, which improves trajectory matching, and thus provides a more robust tracking of objects transitioning between cameras. YOLOv5, ByteTrack, and ResNet50-IBN ReID networks are used for vehicle detection and tracking. Static attributes such as vehicle type and vehicle color are determined from ReID features with SVM. The proposed ReID feature-based attribute categorization shows better performance, than its pure CNN counterpart. Single-camera trajectories (SCTs) are combined into multi-camera trajectories (MCTs) using hierarchical agglomerative clustering (HAC) with time and space constraints (our proposed algorithm is denoted by MCT#MAC). Similarities between SCTs are measured by comparing the mean ReID features cumulated on the trajectory. The system was evaluated on more datasets, and our experiments demonstrate that constraining HAC by manipulating the proximity matrix greatly improves the multi-camera IDF1 score.

Abstract The fast improvement of deep learning methods resulted in breakthroughs in image classification, however, these models are sensitive to adversarial perturbations, which can cause serious problems. Adversarial attacks try to change the model output by adding noise to the input, in our research we propose a combined defense method against it. Two defense approaches have been evolved in the literature, one robustizes the attacked model for higher accuracy, and the other approach detects the adversarial examples. Only very few papers discuss both approaches, thus our aim was to combine them to obtain a more robust model and to examine the combination, in particular the filtering capability of the detector. Our contribution was that the filtering based on the decision of the detector is able to enhance the accuracy, which was theoretically proved. Besides that, we developed a novel defense method called 2N labeling, where we extended the idea of the NULL labeling method. While the NULL labeling suggests only one new class for the adversarial examples, the 2N labeling method suggests twice as much. The novelty of our idea is that a new extended class is assigned to each original class, as the adversarial version of it, thus it assists the detector and robust classifier as well. The 2N labeling method was compared to competitor methods on two test datasets. The results presented that our method surpassed the others, and it can operate with a constant classification performance regardless of the presence or amplitude of adversarial attacks.

Abstract Two questions often arise in the field of the ensemble in multiclass classification problems, (i) how to combine base classifiers and (ii) how to design possible binary classifiers. Error-correcting output codes (ECOC) methods answer these questions, but they focused on only the general goodness of the classifier. The main purpose of our research was to strengthen the bottleneck of the ensemble method, i.e., to minimize the largest values of two types of error ratios in the deep neural network-based classifier. The research was theoretical and experimental, the proposed Min–Max ECOC method suggests a theoretically proven optimal solution, which was verified by experiments on image datasets. The optimal solution was based on the maximization of the lowest value in the Hamming matrix coming from the ECOC matrix. The largest ECOC matrix, the so-called full matrix is always a Min–Max ECOC matrix, but smaller matrices generally do not reach the optimal Hamming distance value, and a recursive construction algorithm was proposed to get closer to it. It is not easy to calculate optimal values for large ECOC matrices, but an interval with upper and lower limits was constructed by two theorems, and they were proved. Convolutional Neural Networks with Min–Max ECOC matrix were tested on four real datasets and compared with OVA (one versus all) and variants of ECOC methods in terms of known and two new indicators. The experimental results show that the suggested method surpasses the others, thus our method is promising in the ensemble learning literature.

Given the challenge of gathering labelled training data for machine learning tasks, active learning has become popular. This paper focuses on the beginning of unsupervised active learning, where there are no labelled data at all. The aim of this zero initialised unsupervised active learning is to select the most informative examples – even from an imbalanced dataset – to be labelled manually. Our solution with proposed selection strategy, called Optimally Balanced Entropy-Based Sampling (OBEBS) reaches a balanced training set at each step to avoid imbalanced problems. Two theorems of the optimal solution for selection strategy are also presented and proved in the paper. At the beginning of the active learning, there is not enough information for supervised machine learning method, thus our selection strategy is based on unsupervised learning (clustering). The cluster membership likelihoods of the items are essential for the algorithm to connect the clusters and the classes, i.e., to find assignment between them. For the best assignment, the Hungarian algorithm is used, and single, multi, and adaptive assignment variants of OBEBS method are developed. Based on generated and real images datasets of handwritten digits, the experimental results show that our method surpasses the state-of-the-art methods.

MOTIVATION: Non-invasive prenatal testing or NIPT is currently among the top researched topic in obstetric care. While the performance of the current state-of-the-art NIPT solutions achieve high sensitivity and specificity, they still struggle with a considerable number of samples that cannot be concluded with certainty. Such uninformative results are often subject to repeated blood sampling and re-analysis, usually after two weeks, and this period may cause a stress to the future mothers as well as increase the overall cost of the test. RESULTS: We propose a supplementary method to traditional z-scores to reduce the number of such uninformative calls. The method is based on a novel analysis of the length profile of circulating cell free DNA which compares the change in such profiles when random-based and length-based elimination of some fragments is performed. The proposed method is not as accurate as the standard z-score; however, our results suggest that combination of these two independent methods correctly resolves a substantial portion of healthy samples with an uninformative result. Additionally, we discuss how the proposed method can be used to identify maternal aberrations, thus reducing the risk of false positive and false negative calls. AVAILABILITY AND IMPLEMENTATION: The open-source code of the proposed methods, together with test data, is freely available for non-commercial users at github web page https://github.com/jbudis/lambda. SUPPLEMENTARY INFORMATION: Supplementary materials are available at Bioinformatics online.