Special Issue: Machine Learning and Data Analysis II

In an age defined by massive data generation, machine learning has emerged as a transformative tool to uncover meaningful patterns and drive intelligent decision-making. By blending statistical methods with algorithmic learning, machine learning enables systems to adapt and improve performance without being explicitly programmed. When coupled with data analysis, the process of inspecting, cleaning, and interpreting data sets, it unlocks powerful capabilities across industries, from predictive healthcare and financial forecasting to personalized marketing and autonomous systems. Together, these technologies form the backbone of modern analytics, redefining how we understand complex information and act upon it.

Following the first volume of this Special Issue, we invited scientists to contribute to its continuation. Finally we received almost 40 submissions and decided to accept 8 of them. The accepted papers present different data analysis techniques:

• Boolean reasoning (contribution 1);
• Classic neural networks (contribution 2);
• Deep learning (contributions 3 and 4);
• Decision tree-based approaches (contribution 4);
• Density estimation (contribution 5);
• Natural language processing (contribution 6);
• Genetic algorithms (contribution 7);
• Clustering (contribution 8).

Moreover, these papers are related to data from completely different domains, such as

• Bioinformatics (contribution 1);
• Biology (contribution 2);
• Chemistry (contributions 3 and 4);
• Text analysis (contribution 6);
• Network traffic (contribution 7);
• Computer vision (contribution 8).

Boolean reasoning [1] is a data analysis paradigm where the original problem is represented as a Boolean formula and the results of its processing correspond directly to the solutions of the initial problem. This approach is widely applied in rough set theory [2] and digital circuit design [3]. Contribution 1 shows the ability of Boolean reasoning to search for a new kind of pattern in biomedical data.

Measurement of blood microflow is crucial to understanding how blood behaves at the cellular level, especially in tiny vessels such as capillaries and arterioles. Tracking changes in microcirculation helps monitor conditions such as liver fibrosis, cardiogenic shock, and rheumatic diseases [4]. Simultaneous measurement of properties such as viscosity, aggregation of red blood cells, and sedimentation reveals how blood flow is altered in disease states [5]. Contribution 2 provides an analysis of the application of artificial neural networks to estimate the velocity of blood microflows.

Chemistry is also a field of data analysis that uses machine learning techniques to interpret experimental results [6]. Notable applications include prediction of polymer properties [7], reaction characteristics [8], and battery behavior [9]. This Special Issue contains two papers that show the application of machine learning in chemistry. Contribution 3 focuses just on battery issues, while contribution 4 applies ML techniques to predict the properties of water–alumina nanofluids.

Density estimation, or, more formally, density function estimation, means the estimation of an unknown density function with some observation-based formula [10]. Kernel-based methods are among the most popular approaches for the estimation of density functions [11,12,13]. The other branch of methods is based on the assumption that empiric observations do not come from a single distribution but are the result of several different ones (normal). This approach is called a Gaussian Mixture Model [14]. Contribution 5 focuses on estimation with neural techniques [15] in the context of normalizing flows.

The late 1950s brought the first suggestion of automated text analysis for the purpose of preparing abstracts from the technical literature [16]. More than 40 years later, a strong mathematical background for so-called “text mining” was already developed [17,18,19]. Today, text mining methods help to cluster texts [20], summarize them [21,22], label them [23], or compare the style of two documents (stylometry) [24,25]. In addition, emotions expressed in the text can become a point of interest for analysis [26]. This last idea was raised in contribution 6. The authors directed their attention towards the change in people’s emotions extracted from online comments over time.

Multi-access Edge Computing (MEC) refers to the provision of cloud computing technology and IT services at the edge of mobile networks [27]. However, due to limited resources in MEC technology, it becomes crucial to ensure a balance, e.g., between maximizing efficiency and minimizing cost (energy consumption) [28]. In contribution 7 the authors present a new algorithm, called GA4QST, which helps to find an optimal solution for the compromise between a satisfactory level of QoS and its potential cost.

Clustering, also called cluster analysis or unsupervised classification, means searching for subgroups of objects or features whose elements are more similar to others within the group than to those outside the group. Such an approach to data analysis was already stated in the 1950s [29]: the k-means algorithm. Since then, a variety of clustering techniques have been developed, for example, DBSCAN [30], Ward’s hierarchical approach [31], and self-organizing maps [32]. However, in some applications, especially in computer vision, when the features are clustered, a specific object (image) can be interpreted from many different points of view. Such datasets are then called multiview or multimodal [33] and require dedicated machine learning techniques [34]. Contribution 8 provides a new solution for this case called MUC: the multi-view utility-based clustering method.