Volumen 31 (2021): Edición 4 (December 2021)
Advanced Machine Learning Techniques in Data Analysis (special section, pp. 549-611), Maciej Kusy, Rafał Scherer, and Adam Krzyżak (Eds.)

Parkinson’s disease (PD) is the second most common neurological disorder in the world. Nowadays, it is estimated that it affects from 2% to 3% of the global population over 65 years old. In clinical environments, a spiral drawing task is performed to help to obtain the disease’s diagnosis. The spiral trajectory differs between people with PD and healthy ones. This paper aims to analyze differences between handmade drawings of PD patients and healthy subjects by applying the SqueezeNet convolutional neural network (CNN) model as a feature extractor, and a support vector machine (SVM) as a classifier. The dataset used for training and testing consists of 514 handwritten draws of Archimedes’ spiral images derived from heterogeneous sources (digital and paper-based), from which 296 correspond to PD patients and 218 to healthy subjects. To extract features using the proposed CNN, a model is trained and 20% of its data is used for testing. Feature extraction results in 512 features, which are used for SVM training and testing, while the performance is compared with that of other machine learning classifiers such as a Gaussian naive Bayes (GNB) classifier (82.61%) and a random forest (RF) (87.38%). The proposed method displays an accuracy of 91.26%, which represents an improvement when compared to pure CNN-based models such as SqueezeNet (85.29%), VGG11 (87.25%), and ResNet (89.22%).

Atanassov’s intuitionistic fuzzy sets and especially his intuitionistic fuzzy relations are tools that make it possible to model effectively imperfect information that we meet in many real-life situations. In this paper, we discuss the new concepts of the transitivity problem of Atanassov’s intuitionistic fuzzy relations in an epistemic aspect. The transitivity property reflects the consistency of a preference relation. Therefore, transitivity is important from the point of view of real problems appearing, e.g., in group decision making in preference procedures. We propose a new type of optimistic and pessimistic transitivity among the alternatives (options) considered and their use in the procedure of ranking the alternatives in a group decision making problem.

The aim of the presented study is to investigate the application of an optimization algorithm based on swarm intelligence to the configuration of a fuzzy flip-flop neural network. Research on solving this problem consists of the following stages. The first one is to analyze the impact of the basic internal parameters of the neural network and the particle swarm optimization (PSO) algorithm. Subsequently, some modifications to the PSO algorithm are investigated. Approximations of trigonometric functions are then adopted as the main task to be performed by the neural network. As a result of the numerical verification of the problem, a set of rules are developed that can be helpful in constructing a fuzzy flip-flop type neural network. The obtained results of the computations significantly simplify the structure of the neural network in relation to similar conditions known from the literature.

One major focus in forensics is the identification of individuals based on different kinds of evidence found at a crime scene and in the digital domain. Here, we assess the potential of using in-vehicle digital data to capture the natural driving behavior of individuals in order to identify them. We formulate a forensic scenario of a hit-and-run car accident with a known and an unknown suspect being the actual driver during the accident. Specific aims of this study are (i) to further develop a workflow for driver identification in digital forensics considering a scenario with an unknown suspect, and (ii) to assess the potential of one-class compared to multi-class classification for this task. The developed workflow demonstrates that in the application of machine learning in digital forensics it is important to decide on the statistical application, data mining or hypothesis testing in advance. Further, multi-class classification is superior to one-class classification in terms of statistical model quality. Using multi-class classification it is possible to contribute to the identification of the driver in the hit-and-run accident in both types of application, data mining and hypothesis testing. Model quality is in the range of already employed methods for forensic identification of individuals.

This work presents an original model for detecting machine tool anomalies and emergency states through operation data processing. The paper is focused on an elastic hierarchical system for effective data reduction and classification, which encompasses several modules. Firstly, principal component analysis (PCA) is used to perform data reduction of many input signals from big data tree topology structures into two signals representing all of them. Then the technique for segmentation of operating machine data based on dynamic time distortion and hierarchical clustering is used to calculate signal accident characteristics using classifiers such as the maximum level change, a signal trend, the variance of residuals, and others. Data segmentation and analysis techniques enable effective and robust detection of operating machine tool anomalies and emergency states due to almost real-time data collection from strategically placed sensors and results collected from previous production cycles. The emergency state detection model described in this paper could be beneficial for improving the production process, increasing production efficiency by detecting and minimizing machine tool error conditions, as well as improving product quality and overall equipment productivity. The proposed model was tested on H-630 and H-50 machine tools in a real production environment of the Tajmac-ZPS company.

The paper studies the output observer design problem for a linear infinite-dimensional control plant modelled as an abstract boundary input/output control system. It is known that such models lead to an equivalent state space description with unbounded control (input) and observation (output) operators. For this class of infinite-dimensional systems we use the Cayley transform to approximate the sophisticated infinite-dimensional continuous-time model by a discrete-time infinite-dimensional one with all involved operators bounded. This significantly simplifies mathematical aspects of the observer design procedure. As is well known, the essential feature of the Cayley transform is that it preserves various system theoretic properties of the control system model, which may be useful in analysis. As an illustration, we consider an example of designing an output observer for the one-dimensional heat equation with measured controls (inputs) in the Neumann boundary conditions, measured outputs in the Dirichlet boundary conditions and an unmeasured output at a fixed point within the domain. Numerical simulations of this example show that the interpolated continuous-time signal, obtained from the discrete-time observer, can be successfully used for tracking the continuous-time plant output.

The divisibility of the second-order minors of the numerators of transfer matrices by their minimal denominators for cyclic fractional linear systems is analyzed. It is shown that all nonzero second-order minors of the numerators of the transfer matrices are divisible by their minimal denominators if and only if the system matrices of fractional standard and descriptor linear systems are cyclic. The theorems are illustrated by examples of fractional standard and descriptor linear systems.

The consensus problem for a class of high-order nonlinear multi-agent systems (MASs) with external disturbance and system uncertainty is studied. We design an online-update radial basis function (RBF) neural network based distributed adaptive control protocol, where the sliding model control method is also applied to eliminate the influence of the external disturbance and system uncertainty. System consensus is verified by using the Lyapunov stability theorem, and sufficient conditions for cooperative uniform ultimately boundedness (CUUB) are also derived. Two simulation examples demonstrate the effectiveness of the proposed method for both homogeneous and heterogeneous MASs.

The paper proposes an algorithm for safeness verification of a Petri net-based specification of the control part of cyber-physical systems. The method involves a linear algebra technique and is based on the computation of the state machine cover of a Petri net. Contrary to the well-known methods, the presented idea does not require obtaining all sequential components, nor the computation of all reachable states in the system. The efficiency and effectiveness of the proposed method have been verified experimentally with a set of 243 test modules (Petri net-based systems). The results of experiments show high efficiency of the proposed method since a solution has been found even for such nets where popular techniques are not able to analyze the safeness of the system. Finally, the presented algorithm is explained in detail using a real-life case-study example of the control part of a cyber-physical system.

Big data, artificial intelligence and the Internet of things (IoT) are still very popular areas in current research and industrial applications. Processing massive amounts of data generated by the IoT and stored in distributed space is not a straightforward task and may cause many problems. During the last few decades, scientists have proposed many interesting approaches to extract information and discover knowledge from data collected in database systems or other sources. We observe a permanent development of machine learning algorithms that support each phase of the data mining process, ensuring achievement of better results than before. Rough set theory (RST) delivers a formal insight into information, knowledge, data reduction, uncertainty, and missing values. This formalism, formulated in the 1980s and developed by several researches, can serve as a theoretical basis and practical background for dealing with ambiguities, data reduction, building ontologies, etc. Moreover, as a mature theory, it has evolved into numerous extensions and has been transformed through various incarnations, which have enriched expressiveness and applicability of the related tools. The main aim of this article is to present an overview of selected applications of RST in big data analysis and processing. Thousands of publications on rough sets have been contributed; therefore, we focus on papers published in the last few years. The applications of RST are considered from two main perspectives: direct use of the RST concepts and tools, and jointly with other approaches, i.e., fuzzy sets, probabilistic concepts, and deep learning. The latter hybrid idea seems to be very promising for developing new methods and related tools as well as extensions of the application area.

This paper considers feature selection as a problem of an aggregation of three state-of-the-art filtration methods: Pearson’s linear correlation coefficient, the ReliefF algorithm and decision trees. A new wrapper method is proposed which, on the basis of a fusion of the above approaches and the performance of a classifier, is capable of creating a distinct, ordered subset of attributes that is optimal based on the criterion of the highest classification accuracy obtainable by a convolutional neural network. The introduced feature selection uses a weighted ranking criterion. In order to evaluate the effectiveness of the solution, the idea is compared with sequential feature selection methods that are widely known and used wrapper approaches. Additionally, to emphasize the need for dimensionality reduction, the results obtained on all attributes are shown. The verification of the outcomes is presented in the classification tasks of repository data sets that are characterized by a high dimensionality. The presented conclusions confirm that it is worth seeking new solutions that are able to provide a better classification result while reducing the number of input features.

The significant benefits of cloud computing (CC) resulted in an explosion of their usage in the last several years. From the security perspective, CC systems have to offer solutions that fulfil international standards and regulations. In this paper, we propose a model for a hash function having a scalable output. The model is based on an artificial neural network trained to mimic the chaotic behaviour of the Mackey–Glass time series. This hashing method can be used for data integrity checking and digital signature generation. It enables constructing cryptographic services according to the user requirements and time constraints due to scalable output. Extensive simulation experiments are conduced to prove its cryptographic strength, including three tests: a bit prediction test, a series test, and a Hamming distance test. Additionally, flexible hashing function performance tests are run using the CloudSim simulator mimicking a cloud with a global scheduler to investigate the possibility of idle time consumption of virtual machines that may be spent on the scalable hashing protocol. The results obtained show that the proposed hashing method can be used for building light cryptographic protocols. It also enables incorporating the integrity checking algorithm that lowers the idle time of virtual machines during batch task processing.

The paper presents the concept of a control unit, i.e., a scenario player, for interactive training pilots in flight simulators. This scenario player is modelled as a hierarchy of finite state machines. Such an approach makes it possible to separate the details of an augmented reality display device which is used in training, from the core module of the system, responsible for contextual organization of the content. Therefore, the first contribution of this paper is the mathematical model of the scenario player as a universal formulation of the self-trained control unit for interactive learning systems, which is applicable in a variety of situations not limited solely to flight simulator related procedures. The second contribution is an experimental verification achieved by extensive simulations of the model, which proves that the proposed approach is capable to properly self-organize details of the context information by tracing preferences of the end users. For that latter purpose, the original algorithm is derived from statistical analysis, including Bayesian inference. The whole approach is illustrated by a real application of training the preflight procedure for the captain of the Boeing 737 aircraft in a flight simulator.

The article presents new tools for investigating the statistical properties of the harmonic signal autocorrelation function (ACF). These tools enable identification of the ACF estimator errors in measurements in which the triggering of the measurements is non-synchronized. This is important because in many measurement situations the initial phase of the measured signal is random. The developed tools enable testing the ACF estimator of a harmonic signal in the presence of Gaussian noise. These are the formulas on the basis of which the statistical properties of the estimator can be determined, including the bias, the variance and the mean squared error (MSE). For comparison, the article also presents the ACF statistical analysis tools used in the conditions of synchronized measurement triggering, known from the literature. Operation of the new tools is verified by simulation and experimental studies. The conducted research shows that differences between the MSE results obtained with the use of the developed formulas and those attained from simulations and experimental tests are not greater than 1 dB.