The rapid development of generative adversarial networks has significantly advanced the generation of synthetic images, presenting valuable opportunities and ethical dilemmas in their potential misuse across various industries. The necessity to distinguish real from AI-generated content is becoming increasingly critical to preserve the integrity of online data. While traditional methods for detecting fake images resulting from image tampering rely on hand-crafted features, the sophistication of manipulated images produced by generative adversarial networks requires more advanced detection approaches. The lightweight approach proposed here is based on convolutional neural networks that comprise only eight convolutional and two hidden layers that effectively differentiate AI-generated images from real ones. The proposed approach was assessed using two benchmark datasets and custom-generated data from Sentinel-2 imagery. It demonstrated superior performance compared to four state-of-the-art methods on the CIFAKE dataset, achieving the highest accuracy of 97.32%, on par with the highest-performing state-of-the-art method. Explainable AI is utilized to enhance our comprehension of the complex processes involved in synthetic image recognition. We have shown that, unlike authentic images, where activations often center around the main object, in synthetic images, activations cluster around the edges of objects, in the background, or in areas with complex textures.

This article introduces a new iterative approach to explainable feature learning. During each iteration, new features are generated, first by applying arithmetic operations on the input set of features. These are then evaluated in terms of probability distribution agreements between values of samples belonging to different classes. Finally, a graph-based approach for feature selection is proposed, which allows for selecting high-quality and uncorrelated features to be used in feature generation during the next iteration. As shown by the results, the proposed method improved the accuracy of all tested classifiers, where the best accuracies were achieved using random forest. In addition, the method turned out to be insensitive to both of the input parameters, while superior performances in comparison to the state of the art were demonstrated on nine out of 15 test sets and achieving comparable results in the others. Finally, we demonstrate the explainability of the learned feature representation for knowledge discovery.

Feature selection is predominantly used in machine learning tasks, such as classification, regression, and clustering. It selects a subset of features (relevant attributes of data points) from a larger set that contributes as optimally as possible to the informativeness of the model. There are exponentially many subsets of a given set, and thus, the exhaustive search approach is only practical for problems with at most a few dozen features. In the past, there have been attempts to reduce the search space using dynamic programming. However, models that consider similarity in pairs of features alongside the quality of individual features do not provide the required optimal substructure. As a result, algorithms, which we will call suboptimal dynamic programming algorithms, find a solution that may deviate significantly from the optimal one. In this paper, we propose an iterative dynamic programming algorithm, which invertsthe order of feature processing in each iteration. Such an alternating approach allows for improving the optimization function by using the score from the previous iteration to estimate the contribution of unprocessed features. The iterative process is proven to converge and terminates when the solution does not change in three successive iterations or when the number of iterations reaches the threshold. Results in more than 95% of tests align with those of the exhaustive search approach, being competitive and often superior to the reference greedy approach. Validation was carried out by comparing the scores of output feature subsets and examining the accuracy of different classifiers learned on these features across nine real-world applications, considering different scenarios with various numbers of features and samples. In the context of feature selection, the proposed algorithm can be characterized as a robust filter method that can improve machine learning models regardless of dataset size. However, we expect that the idea of alternating suboptimal optimization will soon be generalized to tasks beyond feature selection.

The integration of machine learning within educational environments has positioned learning analytics as a valuable approach for enhancing teaching and learning outcomes. This research examines the critical role of feature engineering in learning analytics, where the quality and interpretability of extracted features significantly influence model performance and adoption. Traditional feature engineering methodologies-comprising systematic extraction, evaluation, and selection of relevant features from educational data remain foundational but intensive. While automated feature extraction in the form of feature learning offers scalability, it often lacks interpretability, which is critical in education. Interpretability becomes particularly significant in light of the European Union’s 2024 AI Act, which explicitly mandates transparent and accountable artificial intelligence systems in high-risk domains, including education. Despite growing interest in learning analytics, the field lacks standardized methods for feature engineering, limiting model reproducibility and real-world applicability. Furthermore, highly accurate models often sacrifice interpretability, hindering adoption by educators. This study addresses two core challenges: the absence of reproducible feature engineering methodologies and the need for interpretable models aligned with ethical and legal standards.

Feature selection is predominantly used in machine learning tasks, such as classification and regression. It selects a subset of features (relevant attributes of data points) from a larger set that contributes as optimally as possible to the informativeness of the model. There are exponentially many subsets of a given set, and thus, the exhaustive search is only practical for problems with at most a few dozen features. In the past, there have been attempts to reduce the search space using dynamic programming. However, models that consider similarity in pairs of features besides the quality of individual features do not provide the required optimal substructure. As a result, algorithms, which we will call suboptimal dynamic programming algorithms, find a solution that may deviate significantly from the optimal one. In this paper, we propose a dynamic programming algorithm with an iterative upgrade where the order of feature processing is inverted in each iteration. Such an alternating approach allows for improving the optimization function by using the score from the previous iteration to estimate the contribution of unprocessed features. The iterative process is proven to converge and terminate when the solution does not change in three successive iterations or when the number of iterations reaches the threshold. Results in more than 90 % of tests align with those of the exhaustive search, being competitive and often superior to the reference greedy approach. The validation was done by comparing the scores of output feature subsets and examining the accuracy of different classifiers learned on these features. In the context of feature selection, the proposed algorithm can be characterized as a filter method. However, we expect that the idea of alternating suboptimal optimization will soon be generalized to tasks beyond feature selection.

Student success prediction is a central topic in educational data mining and learning analytics, as institutions increasingly rely on data-driven approaches to enhance learning outcomes. However, the dynamic nature of educational environments raises questions about the long-term reliability of predictive features used in these models. This study aims to investigate the temporal stability of features extracted by sensitivity analysis of predictive models developed by integrating data from various sources, including the e-learning system, student attendance records, teacher opinions, and meteorological data. In this study, the stability of success predictors is modeled using machine learning algorithms – Random Forest and Gradient Boosted Decision Tree. By applying regression metrics, the precision of the model is assessed to determine the reliability of predictive features over time. Identification of the relevant success predictors and their temporal stability rovides insights into significant success predictors in the long term. The results support the development of robust predictive models and highlight key features that contribute to the reliable analysis of student success outcomes.