Viktor Pocajt

PM10 emission forecasting using artificial neural networks and genetic algorithm input variable optimization

This paper describes the development of an artificial neural network (ANN) model for the forecasting of annual PM(10) emissions at the national level, using widely available sustainability and economical/industrial parameters as inputs. The inputs for the model were selected and optimized using a genetic algorithm and the ANN was trained using the following variables: gross domestic product, gross inland energy consumption, incineration of wood, motorization rate, production of paper and paperboard, sawn wood production, production of refined copper, production of aluminum, production of pig iron and production of crude steel. The wide availability of the input parameters used in this model can overcome a lack of data and basic environmental indicators in many countries, which can prevent or seriously impede PM emission forecasting. The model was trained and validated with the data for 26 EU countries for the period from 1999 to 2006. PM(10) emission data, collected through the Convention on Long-range Transboundary Air Pollution - CLRTAP and the EMEP Programme or as emission estimations by the Regional Air Pollution Information and Simulation (RAINS) model, were obtained from Eurostat. The ANN model has shown very good performance and demonstrated that the forecast of PM(10) emission up to two years can be made successfully and accurately. The mean absolute error for two-year PM(10) emission prediction was only 10%, which is more than three times better than the predictions obtained from the conventional multi-linear regression and principal component regression models that were trained and tested using the same datasets and input variables.

The forecasting of municipal waste generation using artificial neural networks and sustainability indicators

The feasibility of modeling municipal waste generation (MWG) for countries at different levels of development using artificial neural networks (ANN) and selected generic indicators of sustainability was investigated. The main goals of this research were to develop ANN-based models for predicting MWG, to overcome the problem of incomplete MWG data, which is notable in developing countries, and to provide a new method for the planning of municipal solid waste management systems as well as for the simulation of various other scenarios. Data from 26 European countries was used in this study as training, test and validation datasets for the developing of ANN models. Since this kind of modeling is particularly important for developing countries where MWG data is missing or incomplete, emphasis was placed on modeling of MWG for Bulgaria and Serbia. Based on a comparison of actual MWG data with predictions given by the model, we show that ANNs can be applied successfully to modeling and forecasting MWG on a national scale. Moreover, the scope for possible application of the model is broad, since it uses generic indicators of sustainability such as gross domestic product, domestic material consumption and resource productivity, and performs well for countries with highly diversified levels of economic development, industrial structure, productivity and output.

Forecasting of Greenhouse Gas Emissions in Serbia Using Artificial Neural Networks

Serbia is attempting to synchronize its development with the basic assumptions of sustainable development and, consequently, data about environmental impact are necessary. The main goal of this study was to investigate and evaluate the possibility of using the artificial neural network technique for predicting the environmental indicators of sustainable development, in order to overcome the problem of incomplete data and to simulate various development scenarios and their environmental impact. Based on the results obtained, it may be concluded that an artificial neural network can be applied to model the greenhouse gas emissions as one of the environmental parameters of sustainable development.

Forecasting human exposure to PM10 at the national level using an artificial neural network approach

A neural network model for predicting country‐level concentrations of the fraction of particulates in the air with sizes less than 10 µm (PM10) has been developed using widely available sustainability and economical/industrial parameters as inputs. The model was trained and validated with the data for 23 European Union (EU) countries plus the EU27 as a group for the period from 2000 to 2008. The inputs for the model were selected using correlation analyses. Country‐level PM10 concentration data that were used as a model output were obtained from the World Bank. The artificial neural network (ANN) model, created with inputs chosen by correlation analyses, has shown very good performance in the forecast of country‐level PM10 concentrations. The mean absolute error for the ANN model prediction, in the case of most of the EU countries, was less than 13%, indicating stable and accurate predictions. The predictions obtained from the principal component regression model, which was trained and tested using the same datasets and input variables, had mean absolute errors from 20% to 150% for most of the countries. The wide availability of input parameters used in this model can overcome the problem of lack and scarcity of data in many countries, which can in turn prevent the determination of human exposure to PM10 at the national level. Copyright

Using data mining to improve digital library services

Purpose – This paper aims to propose a solution for recommending digital library services based on data mining techniques (clustering and predictive classification).Design/methodology/approach – Data mining techniques are used to recommend digital library services based on the users profile and search history. First, similar users were clustered together, based on their profiles and search behavior. Then predictive classification for recommending appropriate services to them was used. It has been shown that users in the same cluster have a high probability of accepting similar services or their patterns.Findings – The results indicate that k‐means clustering and Naive Bayes classification may be used to improve the accuracy of service recommendation. The overall accuracy is satisfying, while average accuracy depends on the specific service. The results were better for frequently occurring services.Research limitations/implications – Datasets were used from the KOBSON digital library. Only clustering and ...

Concentrations of selected trace elements in mineral and spring bottled waters on the Serbian market

Eight selected trace elements, which are generally included in regulations, were analyzed in 23 types of bottled waters. Ten mineral and seven spring bottled waters were from the Serbian market and six mineral bottled waters were obtained in different EU countries. For the purpose of comparison, selected tap waters were also analyzed. Inductively coupled plasma mass spectrometry (ICP-MS) was used for the analysis of trace elements (arsenic, cadmium, copper, manganese, nickel, lead and antimony). Results were compared with the Serbian regulations for bottled water, EU regulations and guideline values set by the World Health Organization for drinking water. With few exceptions, the trace element levels of most bottled waters were below the guideline values. However, a higher content of antimony was observed in waters from polyethylene terephthalate (PET) containers, indicating a potential leaching of this element from the plastic packaging.

Prediction of nitrogen oxides emissions at the national level based on optimized artificial neural network model

Nitrogen oxides (NOx) emissions into the atmosphere have multiple negative effects on the environment and effects directly and indirectly on human health. This paper describes the development of a model for NOx emission prediction at the national level based on artificial neural networks (ANNs) and on widely available sustainability, industrial, and economical parameters as input variables. In this study, 11 sustainability, industrial, and economical parameters were chosen as potential input variables. The ANN models were trained, validated, and tested with available data for 17 European countries, USA, China, Japan, Russia, and India for the years 2001 to 2008. The ANN modeling was performed using general regression neural network (GRNN), and correlation and variance inflation factor (VIF) analysis were applied to reduce the number of input variables. The best results were obtained using the selection of inputs based on the correlation between input variables, which provided a more accurate prediction than the GRNN model created with all initial selected input variables. Sensitivity analysis showed that the input variables with the largest influences on the GRNN model results were (in descending order) electricity production from oil sources, agricultural land, fossil fuel energy consumption, number of vehicles, gross domestic product, energy use, and electricity production from coal sources.

Prediction of clearing temperatures of bent-core liquid crystals using decision trees and multivariate adaptive regression splines

ABSTRACT Accurate prediction of transition temperature is very helpful for the design of new liquid crystals (LCs) because even small changes in structure can dramatically alter the transition temperature, and therefore the synthesis of LCs should not be governed only by chemical intuition. A quantitative structure–property relationship (QSPR) study was performed on 243 five-ring bent-core LCs in order to predict their clearing temperatures using molecular descriptors. Decision tree and multivariate adaptive regression splines (MARS), techniques well suited for high-dimensional data analysis, were applied to select important descriptors (dimension reduction) and to generate nonlinear models. These techniques were applied both on two-dimensional (2D) descriptors only and on the pool of 2D and 3D descriptors (2&3D). The obtained QSPR models were tested using 15% of available data, and their performance and ability to generalise were analysed using multiple statistical metrics. The best results for the external test set were obtained using the MARS model created with 2&3D descriptors, with a high correlation coefficient of r = 0.95 and a root mean squared error of 7.41 K. All metrics suggest that the proposed QSPR model, generated by MARS, is a robust and satisfactorily accurate approach for the prediction of clearing temperatures of bent-core LCs. GRAPHICAL ABSTRACT

Modeling the BOD of Danube River in Serbia using spatial, temporal, and input variables optimized artificial neural network models

This paper describes the application of artificial neural network models for the prediction of biological oxygen demand (BOD) levels in the Danube River. Eighteen regularly monitored water quality parameters at 17 stations on the river stretch passing through Serbia were used as input variables. The optimization of the model was performed in three consecutive steps: firstly, the spatial influence of a monitoring station was examined; secondly, the monitoring period necessary to reach satisfactory performance was determined; and lastly, correlation analysis was applied to evaluate the relationship among water quality parameters. Root-mean-square error (RMSE) was used to evaluate model performance in the first two steps, whereas in the last step, multiple statistical indicators of performance were utilized. As a result, two optimized models were developed, a general regression neural network model (labeled GRNN-1) that covers the monitoring stations from the Danube inflow to the city of Novi Sad and a GRNN model (labeled GRNN-2) that covers the stations from the city of Novi Sad to the border with Romania. Both models demonstrated good agreement between the predicted and actually observed BOD values.

A QSPR study on the liquid crystallinity of five-ring bent-core molecules using decision trees, MARS and artificial neural networks

Accelerating progress in the discovery of new bent-core liquid crystal (LC) materials with enhanced features relies on the understanding of structure–property relationships that underline the formation of LC phases. The aim of this study was to develop a model for the prediction of LC behaviour of five-ring bent-core systems using a QSPR approach that combines dimension reduction techniques (e.g. genetic algorithms etc.) for the selection of molecular descriptors and decision trees, multivariate adaptive regression splines (MARS) and artificial neural networks (ANN) as classification methods. A total of 27 models based on separate pools of calculated molecular descriptors (2D; 2D and 3D) and published experimental outcomes were evaluated. Overall, the results suggest that the acquired ANN LC classifiers are usable for the prediction of LC behaviour. The best of these models showed high accuracy and precision (91% and 97%). Since the best classifier is able to successfully capture trends in a homologous series, it can be used not only to screen new bent-core structures for potential LCs, but also for the estimation of influence of structural modifications on LC phase formation, as well as for the evaluation of LC phase stability.