People
doc. Ing. Evženie Uglickich, CSc.

2025, Journal of Computational Science, 85, ISSN 1877-7511

In this paper, we address the task of modeling and predicting count data, with an application to traffic counts on selected urban roads in Prague. We investigated the relationship between multiple counts, designating one of them as the target variable (e.g., data from a key road section) and the others as explanatory counts. Defining traffic count data as the number of vehicles passing through a selected road section per unit of time, we use a framework based on Poisson models to develop a progressive methodology, which we compared with existing models. Working with multimodal count data, we propose the following main steps for the methodology: (i) cluster analysis of explanatory counts using recursive Bayesian estimation of Poisson mixtures; (ii) target count model estimation via local Poisson regressions at identified locations, capturing local relationships between target and explanatory counts; and (iii) prediction of target counts through real-time location detection. The algorithm’s properties were first investigated using simulated data and then validated with real traffic counts. Experimental results indicate that the proposed algorithm outperforms classical Poisson and negative binomial regressions, decision tree and random forest classifiers, as well as a multi-layer perceptron, in predicting traffic count data across various quality metrics, even for weakly correlated data. Applied to traffic count data, the promising performance demonstrated by the proposed algorithm offers an optimistic vision for traffic prediction and urban planning, suggesting its potential as a valuable tool for enhancing transportation efficiency by optimizing the timing of city traffic lights to improve traffic flow.

2024, 2024 Smart City Symposium Prague (SCSP), New York, IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC), p. 1-5), ISBN 979-8-3503-6095-0, ISSN 2691-3666

The paper deals with a comparative analysis of three widely used data analysis methods: logistic regression, random forest, and neural networks. These methods have been evaluated in terms of accuracy, and computational efficiency and applied to different types of data sets, including both simulated and real MaaS data. The study aims to compare the efficiency of each method in classification tasks. The study leads to specific recommendations on which method to use under various circumstances, contributing to the decision-making process in data analysis projects. We have shown that random forests generally provide better accuracy and are resistant to over-training. Neural networks can achieve comparable performance under certain conditions, although at a high computational cost. Logistic regression shows limitations in dealing with complex data structures.

2024

This data was provided by the mobile measuring laboratory MobiLab from the Czech Technical University in Prague (https://mobilab.fd.cvut.cz). It presents four data sets with traffic counts measured every minute over a 24-hour period at the following four selected areas in Prague: Stodůlky (September 2021), Barrandov (June 2022), Radlice (September 2022), and Velká Chuchle (October 2022). At each location, 3 to 6 points were chosen where traffic counts were collected in all possible directions along the corresponding road section or intersection. Directions are denoted by numbers 1-4 in the columns of each measured point. This data will be used in the research paper, where detailed information about the traffic counts will be available. References to the paper will be added later.

2024, Neural Network World, 34 (2), p. 89-110), ISSN 2336-4335

This paper deals with the analysis of the relationship between locations and types of crime observed in the Czech Republic. Cluster analysis of crime data based on the recursive Bayesian mixture estimation algorithm is used to identify crime hotspots and estimate local models of crime type. The experiments report that the 2D configuration of the algorithm allows the detection of crime hotspots online. The 3D configuration provides 29% more accurate crime type models than 2D clustering and alternative data mining algorithms. For the data set used, it was determined in which crime hotspots the most serious and most frequent types of crime can be expected to occur with the highest probability. The limitation of the study is the artificial support of the 3D clusters by the fully continuous data vector with the recoded values of the crime type. The potential use of the algorithm is expected in online web applications for sharing information on criminal offenses managed by the Police of the Czech Republic with the public and local government entities in the Czech Republic.

2023, 2023 Smart City Symposium Prague, New York, IEEE Press), p. 1-6), ISBN 979-8-3503-2162-3, ISSN 2691-3666

The paper deals with the analysis of histograms of discrete data collected in questionnaires obtained for individual realizations of the target variable. The main aim of the analysis isto explore the influence of combinations of explanatory variables, represented by responses to the questionnaire, on the behavior of the target variable of the questionnaire. In this paper, an automated approach to histogram comparison is proposed based on coding combinations of data and detecting significant differences in frequencies using the Marascuilo procedure. This is the main contribution of the paper. The approach is validated using a simulated questionnaire in which respondents answered regarding their intention to purchase an electric vehicle subject to finance, leasing, and charging availability, as well as their driving style. The results of the experiments are demonstrated.

2023, Proceedings of the The 12th IEEE International Conference on Intelligent Data Acquisition and Advanced Computing Systems: Technology and Applications (IDAACS) IDAACS’2023, Dortmund, IEEE), p. 51-56), ISBN 979-8-3503-5804-9, ISSN 2770-4254

The paper considers the problem of online prediction of a count variable based on real-time explanatory data of mixed count and categorical nature. The presented solution is based on (i) recursive Bayesian estimation of a mixture model of Poisson-distributed explanatory counts, using the categorical explanatory variable as a measurable pointer of the mixture, (ii) construction of a mixture of local Poisson regressions on the clustered data, and (iii) use of the pre-estimated mixtures for online prediction of the target count using actual measured explanatory data. The latter is one of the main contributions of the proposed approach. In addition, the dynamic model of the categorical explanatory variable preserves the functionality of the algorithm in case of its measurement failure. The experiments with simulations and real data report lower prediction errors compared to theoretical counterparts.

2023, Informatics in Control, Automation and Robotics. ICINCO 2021. Lecture Notes in Electrical Engineering, vol 1006, Springer, Cham), p. 163-184), ISBN 978-3-031-26474-0, ISSN 1876-1119

The chapter focuses on the description of the relationship of the count variable and explanatory Gaussian variables. The cluster-based model is proposed, which is constructed on conditionally independent Gaussian clusters captured in real time using recursive algorithms of the Bayesian mixture estimation theory. The resulting model is expected to be used for predicting count data using real time Gaussian observations. The Poisson distribution of the count data is used as a basic model. However, in reality, count data often do not satisfy the Poisson assumption of equal mean and variance. For this case, five cluster-based Poisson-related models of overdispersed data have been studied. The experimental part of the chapter demonstrates a comparison of the prediction accuracy of the considered models with two theoretical counterparts for the case of weak and strong overdispersion with the help of simulations. The paper reports that the most accurate prediction in average has been provided by the cluster-based Generalized Poisson models.

2023, Neural Network World, 33 (4), p. 291-315), ISSN 2336-4335

The development of traffic state prediction algorithms embedded in intelligent transportation systems is of great importance for improving traffic conditions for drivers and pedestrians. Despite the large number of prediction methods, existing limitations still confirm the need to find a systematic solution and its adaptation to specific traffic data. This paper focuses on the relationship between traffic flow states in different urban locations, where these states are identified as clusters of traffic counts. Extending the recursive Bayesian mixture estimation theory to the Poisson mixtures, the paper uses the mixture pointers to construct the traffic state prediction model. Using the predictive model, the cluster at the target urban location is predicted based on the traffic counts measured in real time at the explanatory urban location. The main contributions of this study are: (i) recursive identification and prediction of the traffic state at each time instant, (ii) straightforward Poisson mixture initialization, and (iii) systematic theoretical background of the prediction approach. Results of testing the prediction algorithm on real traffic counts are presented.

2022, Neural Network World, 32 (2), p. 73-92), ISSN 1210-0552

The present paper describes hand detection application implemented on Xilinx Zynq Ultrascale+ device, comprising multi-core processor ARM Cortex A53 and FPGA programmable logic. It uses ultrasound data and is based on adaptive QRD RLS lattice algorithm extended with hypothesis testing. The algorithm chooses between two use-cases: (1) ''there is a hand in front of the device'' vs (2) ''there is no hand in front of the device''. For these purposes a new structure of the identification models was designed. The model presenting use-case (1) is a regression model, which has the order sufficient to cover all incoming data. The model responsible for use-case (2) is a regression model, which has a smaller order than the model (1) and a certain time delay, covering the maximal distance where the hand can possibly appear. The offered concept was successfully verified using real ultrasound data in MATLAB optimized for parallel processing and implemented in parallel on four cores of ARM Cortex A53 processor. It was proved that computational time of the algorithm is sufficient for applications requiring real-time processing.

2022, Neural Network World, 32 (1), p. 15-41), ISSN 1210-0552

The paper deals with the task of modeling discrete questionnaire data with a reduced dimension of the model. The discrete model dimension is reduced using the construction of local models based on independent binomial mixtures estimated with the help of recursive Bayesian algorithms in the combination with the naive Bayes technique. The main contribution of the paper is a three-phase algorithm of the discrete model dimension reduction, which allows to model high-dimensional questionnaire data with high number of explanatory variables and their possible realizations. The proposed general solution is applied to the traffic accident questionnaire analysis, where it takes the form of the classification of the accident circumstances and prediction of the traffic accident severity using the currently measured discrete data. Results of testing the obtained model on real data and comparison with theoretical counterparts are demonstrated.

2021, Proceedings of the 18th International Conference on Informatics in Control, Automation and Robotics - ICINCO, Setùbal, SciTePress), p. 641-648), ISBN 978-989-758-522-7, ISSN 2184-2809

This paper aims at presenting the on-line non-iterative form of Bayesian mixture estimation. The model used is composed of a set of sub-models (components) and an estimated pointer variable that currently indicates the active component. The estimation is built on an approximated Bayes rule using weighted measured data. The weights are derived from the so called proximity of measured data entries to individual components. The basis for the generation of the weights are integrated likelihood functions with the inserted point estimates of the component parameters. One of the main advantages of the presented data analysis method is a possibility of a simple incorporation of the available prior knowledge. Simple examples with a programming code as well as results of experiments with real data are demonstrated. The main goal of this paper is to provide clear description of the Bayesian estimation method based on the approximated likelihood functions, called proximities.

2021, Proceedings of the 18th International Conference on Informatics in Control, Automation and Robotics - ICINCO, Setùbal, SciTePress), p. 600-608), ISBN 978-989-758-522-7, ISSN 2184-2809

The paper deals with predicting a discrete target variable described by the Poisson distribution based on the discretized Gaussian explanatory data under condition of the multimodality of a system observed. The discretization is performed using the recursive mixture-based clustering algorithms under Bayesian methodology. The proposed approach allows to estimate the Gaussian and Poisson models existing for each discretization interval of explanatory data and use them for the prediction. The main contributions of the approach include: (i) modeling the Poisson variable based on the cluster analysis of explanatory continuous data, (ii) the discretization approach based on recursive mixture estimation theory, (iii) the online prediction of the Poisson variable based on available Gaussian data discretized in real time. Results of illustrative experiments and comparison with the Poisson regression is demonstrated.

2020, Proceedings of the IEEE 14th International Symposium on Applied Computational Intelligence and Informatics SACI 2020, Budapest, IEEE Hungary Section, University Obuda), p. 77-82), ISBN 978-1-7281-7377-1

The paper deals with the task of modeling mixed continuous Gaussian and discrete Poisson data observed on a multimodal system. The proposed solution is based on recursive algorithms of Bayesian mixture estimation. The main contributions of the approach are: (i) the use of the discretized information of normal variables in the form of their clusters in order to keep the one-pass recursive estimation methodology and (ii) the prediction of the multimodal Poisson variable. Experiments with simulated and real data are presented.

2020, 2020 IEEE 24th International Conference on Intelligent Engineering Systems (INES), Budapest, IEEE Hungary Section, University Obuda), p. 67-72), ISBN 978-1-7281-1059-2, ISSN 1543-9259

The paper deals with modeling ordinal discrete random variables with a high number of nonnegative realizations. The prediction of the Rayleigh distribution learned on clusters of the explanatory variables is proposed. The proposed solution consists of the clustering and estimation phases based on the knowledge both of the target and explanatory variables, and the prediction phase using only the information from the explanatory variables. The main contributions of the approach are: (i) using the discretized knowledge of clusters of the explanatory variables and (ii) describing nonnegative discrete data by the multimodal Rayleigh distribution. Experiments with a data set from a tram network are provided.

2019, METRON, 77 (3), p. 253-270), ISSN 0026-1424

The paper focuses on a problem of comparing clusterings with the same number of clusters obtained as a result of using different clustering algorithms. It proposes a method of the evaluation of the agreement of clusterings based on the combination of the Cohen's kappa statistic and the normalized mutual information. The main contributions of the proposed approach are: (i) the reliable use in practice in the case of a small fixed number of clusters, (ii) the suitability to comparing clusterings with a higher number of clusters in contrast with the original statistics, (iii) the independence on size of the data set and shape of clusters. Results of the experimental validation of the proposed statistic using both simulations and real data sets as well as the comparison with the theoretical counterparts are demonstrated.

2019, Proceedings of the 16th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2019), Madeira, SciTePress), p. 617-624), ISBN 978-989-758-380-3

The paper deals with the problem of modeling the passenger demand in the tram transportation network. The passenger demand on the individual tram stops is naturally influenced by the number of boarding and disembarking passengers, whose measuring is expensive and therefore they should be modeled and predicted. A mixture of Poisson components with the dynamic pointer estimated by recursive Bayesian estimation algorithms is used to describe the mentioned variables, while their prediction is solved with the help of the Poisson regression. The main contributions of the presented approach are: (i) the model of the number of boarding and disembarking passengers. (ii) the real-time data incorporation into the model. (iii) the recursive estimation algorithm with the normal approximation of the proximity function. The results of experiments with real data and the comparison with theoretical counterparts are demonstrated.

2019, Informatics in Control, Automation and Robotics. ICINCO 2017. Lecture Notes in Electrical Engineering., Springer, Cham), p. 679-698), ISBN 978-3-030-11292-9, ISSN 1876-1119

The paper provides a practical guide on initialization of the recursive mixture-based clustering of non-negative data. For modeling the non-negative data, mixtures of uniform, exponential, gamma and other distributions can be used. Initialization is known to be an important task for a start of the mixture estimation algorithm. Within the considered recursive approach, the key point of initialization is a choice of initial statistics of the involved prior distributions. The paper describes several initialization techniques for the mentioned types of components that can be beneficial primarily from a practical point of view.

2019, Transportation Research Part B: Methodological, 128, p. 254-270), ISSN 0191-2615

This paper deals with the task of modeling the driving style depending on the driving environment. The model of the driving style is represented as a two-layer mixture of normal components describing data with two pointers: outer and inner. The inner pointer indicates the actual driving environment categorized as “urban”, “rural” and “highway”. The outer pointer through the determined environment estimates the active driving style from a fuel economy point of view as “low consumption”, “middle consumption” and “high consumption”. All of these driving styles are assumed to exist within each driving environment due to the two-layer model. Parameters of the model and the driving style are estimated online, i.e., while driving using a recursive algorithm under the Bayesian methodology. The main contributions of the presented approach are: (i) the driving style recognition within each of urban, rural and highway environments as well as in the case of switching among them. (ii) the two-layer pointer, which allows us to incorporate the information from continuous data into the model. (iii) the potential use of the data-based model for other measurements using corresponding distributions. The approach was tested using real data.

2018, Practical Issues of Intelligent Innovations. Studies in Systems, Decision and Control, Springer, Cham), p. 313-330)

2018, Intuitionistic Fuzziness and Other Intelligent Theories and Their Applications. Studies in Computational Intelligence., Springer, Cham), ISBN 978-3-319-78931-6, ISSN 1860-9503

The initialization is known to be a critical task for running a mixture estimation algorithm. A majority of approaches existing in the literature are related to initialization of the expectation-maximization algorithm widely used in this area. This study focuses on the initialization of the recursive mixture estimation for the case of normal components, where the mentioned methods are not applicable. Its key part is a choice of the initial statistics of normal components.

2017, ISBN 978-3-319-64670-1, ISSN 2191-544X

This book provides a general theoretical background for constructing the recursive Bayesian estimation algorithms for mixture models. It collects the recursive algorithms for estimating dynamic mixtures of various distributions and brings them in the unified form, providing a scheme for constructing the estimation algorithm for a mixture of components modeled by distributions with reproducible statistics. It offers the recursive estimation of dynamic mixtures, which are free of iterative processes and close to analytical solutions as much as possible. In addition, these methods can be used online and simultaneously perform learning, which improves their efficiency during estimation. The book includes detailed program codes for solving the presented theoretical tasks. Codes are implemented in the open source platform for engineering computations. The program codes given serve to illustrate the theory and demonstrate the work of the included algorithms.

2017, Transportation Research Part C: Emerging Technologies, p. 23-36), ISSN 0968-090X

The paper focuses on a task of stochastic modeling the driving style and its online estimation while driving. The driving style is modeled by means of a mixture model with normal and categorical components as well as a data-dependent pointer. The main contributions of the presented approach are: (i) the online estimation of the driving style while driving, taking into account data up to the current time instant; (ii) the joint model for continuous and discrete data measured on a vehicle; (iii) the data-dependent model of the driving style conditioned by the values of fuel consumption; (iv) the use of the model both for detection of clusters according to the driving style and prediction of the fuel consumption along with other variables; and (v) the universal modeling with the help of mixtures, which allows us to use different combinations of components and pointer models as well as to specify the initialization approach suitable for the considered problem.

2017, Intelligent Systems and Informatics (SISY), 2017 IEEE 15th International Symposium on, Budapest, IEEE Hungary Section, University Obuda), ISBN 978-1-5386-3855-2, ISSN 1949-0488

This paper deals with a modeling of data by several mixtures of different distributions within a task of clustering. This issue can be required from a practical point of view, e.g., for a multi-modal system, which generates measurements described by different distributions. The approach is based on the partition of the data on several parts, the factorization of the joint probability density function according to these parts and the estimation of each conditional mixture separately. Due to the data-based construction of the general model from the estimated components, the most suitable combination of the components is used at each time instant.

2017, Proceedings of the 14th International Conference on Informatics in Control, Automation and Robotics, Madeira, SciTePress), ISBN 978-989-758-263-9

The paper deals with a task of initialization of the recursive mixture estimation for the case of uniform components. This task is significant as a part of mixture-based clustering, where data clusters are described by the uniform distributions. The issue is extensively explored for normal components. However, sometimes the assumption of normality is not suitable or limits potential application areas (e.g., in the case of data with fixed bounds). The use of uniform components can be beneficial for these cases. Initialization is always a critical task of the mixture estimation. Within the considered recursive estimation algorithm the key point of its initialization is a choice of initial statistics of components. The paper explores several initialization approaches and compares results of clustering with a theoretical counterpart. Experiments with real data are demonstrated.

2017, Proceedings of International Conference on Intelligent Informatics and BioMedical Sciences ICIIBMS 2017, Okinawa, Okinawa Institute of Science and Technology), p. 63-70), ISBN 978-1-5090-6664-3, ISSN 2189-8723

The paper deals with the mixture-based clustering of anonymized data of patients with leukemia. The presented clustering algorithm is based on the recursive Bayesian mixture estimation for the case of exponential components and the data-dependent dynamic pointer model. The main contribution of the paper is the online performance of clustering, which allows us to actualize the statistics of components and the pointer model with each new measurement. Results of the application of the algorithm to the clustering of hematological data are demonstrated and compared with theoretical counterparts.