Common Predictor Variables Research Articles

The effect of electricity consumption determinants in household load forecasting models

Usually, household electricity consumption fluctuates, often driven by several electrical consumption determinants such as income, household size, and price. Recently, research studies on the investigation of predictor variables in household electricity consumption have increased especially in the developing and newly industrialized countries. However, the studies just focus on identifying the predictor variables of household electricity consumption that influence load forecasting models. In Tanzania, for instance, scholars found that using the “income” determinant improves the performance of a forecasting model. The scholars suggest without any empirical bases that adding more predictor variables would have improved the accuracy of the model. This study aims to analyze the effect of the number of predictor variables on household load forecasting performance based on Tanzania’s data. Nonlinear regression based on a Weibull function and multivariate adaptive regression splines approaches are used for this purpose. Our findings indicate that income, household size, and number of appliances are common predictor variables of household consumption in developing countries. The measured forecasting root-mean-square error (RMSE) when using income, household size, and the number of appliances is 0.8244, 1.2314, and 0.9868, respectively. Finally, we forecasted load using all three determinants and the RMSE dropped to 0.7031. Having obtained the smaller value of RMSE when all predictors are used reveals that the inclusion of all three predictor variables in load forecasting leads to a significant decrease in RMSE by 14.73%. Therefore, the study recommends using multiple predictor variables in load forecasting models to increase accuracy.

Organizational innovation and business model innovation: bridges from a systematic literature review

PurposeThe purpose of this paper is to create a model that connects drivers between organizational innovation and business model innovation (BMI) to guide empirical research and the design of innovation management strategies.Design/methodology/approachThe model was designed based on the results of a systematic literature review over the past 25 years that provides common predictor variables to build bridges between these two types of innovations.FindingsIt is a conceptual relationship between organizational innovation and BMI based on processes, new structures and customer relationship management. Moreover, there are five bridges from common predictors: strategy, top management, exploratory learning, technological innovation and environmental complexity.Originality/valueThe relationships between organizational innovation and BMI have been neglected in the literature. The model fills this gap by proposing hypotheses for empirical research and critical variables and relationships to steer organizational and business model innovation.

Households condition and satisfaction towards post-disaster resettlement: The case of typhoon Haiyan resettlement areas in Tacloban City

This paper provides an assessment of residential satisfaction of post-disaster resettlement areas that were built in Tacloban City, Leyte, Philippines after Typhoon Haiyan. We analyze four components of residential satisfaction: environmental, infrastructure and management, social and neighborhood and economic-livelihood services. Results from the study indicate that the residents are moderately satisfied with the post-disaster housing provided after typhoon Yolanda. A multiple linear regression was employed in the study to determine the factors that are correlated with housing satisfaction. Results from the regression show that use of houses for cultural activities, presence of health infrastructure and membership in associations were the common predictor variables to determine residential satisfaction. Meanwhile, the presence of infrastructural defects such as wall cracks and distance from the city center are negatively associated with housing satisfaction. The data from this study provides specific factors that can influence residential satisfaction which can provide targeted interventions for future resettlement projects.

Measuring a cross-sectional sample of students’ intentions to engage with science and modelling associations according to two theoretical perspectives

ABSTRACT Social cognitive career theory (SCCT) and the theories of reasoned action and planned behaviour (TRAPB) have both guided studies seeking to explain students’ intentions and decisions related to future science engagement. The purpose of this study was to simultaneously validate measures for these two social psychological models and closely examine these constructs and their associations. Students were sampled in two groups, across grades 5 through 10, from rural schools in the United States (N = 538). Each group was administered two sets of measures, and each set of measures was associated with a theoretical model, and included corresponding sub-scales suitable for self-report. Respondents also supplied information about a common set of predictor variables based on a review of the literature. Structural equation modelling was used to examine the constructs more closely. Results from confirmatory factor analysis were generally consistent with the theoretical structures. Multiple indicators multiple causes (MIMIC) models were generated to compare SCCT and TRAPB, including associations between factors and variables. Associations between common predictor variables like race and gender differed between the MIMIC models. Frequency of talk with family members, perceived science ability, gender, and race were all significantly associated with multiple factors in both the SCCT- and TRAPB-aligned models.

Using habitat suitability information from across a species range to inform understudied ecosystems

AbstractKnowledge of habitat suitability (HS) is required for effective conservation planning, yet it can be difficult to obtain. In the absence of local HS information, managers often use studies from other areas to guide their decisions, typically without local validation. We tested the hypothesis that HS is consistent across a species range, and thus reliably extrapolated to understudied sites. We focused on the puma, Puma concolor, a widely distributed large carnivore of conservation and management importance. We conducted a meta-analysis of 41 studies to calculate a mean effect size for six of the most common predictor variables used in mammal HS research. Using the estimated effect sizes in a regression model which included all six variables, we created a new HS model for pumas in an understudied site, the Tumbesian Region (TR). We contrasted predictions from this range-wide model with those from three more regionally specific HS models: a model developed in the Caatinga, Brazil (Caatinga model), a model considering only tropical studies (tropical model), and a model using only studies from the United States and Canada (temperate model). We used puma detection rates from camera trap surveys across the TR to validate model predictions. Although mean effect sizes of habitat predictor variables varied across puma range, all models provided useful predictions of HS for pumas in the TR (area under the receiver operating characteristic curve [AUC] > 0.64). Unexpectedly, the temperate model was best at predicting puma HS in the TR (AUC: 0.77; rs = 0.3), followed by the range-wide model (AUC = 0.73; rs = 0.29). The tropical and Caatinga models had lower predictive accuracy (AUC = 0.68; rs = 0.28 and AUC = 0.64; rs = 0.23, respectively). The accuracy of the tropical model improved when the area of potential recent puma extirpation was excluded from the validation data set. These results highlight that although HS for P. concolor varies across the species range, information collected across a wide range of sites may be better than only locally or regionally specific information for informing HS in understudied sites (e.g., for habitat protection, restoration areas). Given the pressing need for actions to address widespread biodiversity declines, existing knowledge can be used to predict HS to data-poor regions and inform conservation planning while also motivating model validations and targeted data collection.

Some aspects of response variable selection and estimation in multivariate linear regression

Multivariate linear regression analysis is an important technique for modeling the predictive relationships of multiple related response variables on a set of common predictor variables. Numerous studies have been conducted on situations where response variables are given and only predictor variables are subject to variable selection. In practice, however, some response variables do not depend on any of the predictor variables and have very small regression coefficients, implying that response variables need to be selected. Several methods have been proposed for response variable selection in multivariate linear regression. Examples include Bonferroni selection, linear step-up selection, adaptive linear step-up selection, multiple-stage linear step-up selection, response best-subset selection and sparse envelope selection. In this article, we address some aspects of response variable selection focusing on the above-mentioned examples concerning methodological developments, theoretical properties and computational algorithms. We address their performances under the recall rate or true positive rate, true negative rate, precision rate, F-measure, model size and their standard deviations via simulation studies. We also highlight two issues that require further study.

Descriptive and Kinetic Analysis of Two Different Vertical Jump Tests Among Youth and Adolescent Male Basketball Athletes Using a Supervised Machine Learning Approach.

Gillett, J, De Witt, J, Stahl, CA, Martinez, D, and Dawes, JJ. Descriptive and kinetic analysis of two different vertical jump tests among youth and adolescent male basketball athletes using a supervised machine learning approach. J Strength Cond Res 35(10): 2762-2768, 2021-The countermovement jump (CMJ) is a functional movement in basketball and is also frequently used as an assessment of lower-body power. The CMJ can be performed in a variety of manners, and multiple variables can be extracted, and calculated, from the ground reaction force (GRF) time curve. The purpose of this article is to present kinematic and kinetic data collected from adolescent male basketball players during performance of the CMJ with hands on hips (HOH) or with an arm swing while reaching overhead to a target (i.e., vertical jump reach [VJR]). This study also sought to determine the effectiveness of a machine learning algorithm to identify the most important features that relate to jump height. Bilateral GRF data were collected on 89 right-handed male basketball athletes (age: 13.19 ± 0.72 year old, mass: 60.44 ± 13.35 kg, standing reach height: 228.49 ± 16.79 cm) using force platforms (Forcedecks, Vald Performance, Newstead, Queensland, Australia) and their associated software. Fifty-six bilateral kinematic and kinetic variables from each condition were analyzed using supervised machine learning to identify the top 10 important features to predict jump height in each condition, and to predict VJR height using HOH data. Vertical center of mass flight height was greater during VJR trials than during HOH trials (38.9 ± 6.8 cm vs. 32.6 ± 5.5 cm, respectively). The only common predictor variables between the conditions were concentric impulse and peak power. HOH jump data were able to predict VJR height with a mean error of 7.13 cm. These data suggest that important force platform data relating to jump height differ depending on test condition, and that data from CMJ performed with HOH, particularly peak power, concentric impulse, and concentric rate of power development, can be used to predict jump height during functional performance.

Ground-motion models for Vrancea intermediate-depth earthquakes

A newly compiled high-quality ground-shaking dataset of 207 intermediate-depth earthquakes recorded in the Vrancea region of the south-eastern Carpathian mountains in Romania was used to develop region-specific empirical predictive equations for various intensity measures: peak ground acceleration, peak ground velocity, and 5%-damped pseudo-spectral acceleration up to 10 s. Besides common predictor variables (e.g. moment magnitude, depth, hypocentral distance, and site conditions), additional distance scaling parameters were added to describe the specific attenuation pattern observed at the stations located not only on the back and fore but also along the Carpathian arc. In this model, we introduce a proxy measure for the site as the fundamental frequency of resonance to characterize the site response at each recording seismic station beside the soil classes. To additionally reduce the site-to-site variability, a non-ergodic methodology was considered, resulting in a lower standard deviation of about 25%. Statistical evaluation of the newly proposed ground-motion models indicates robust performance compared to regional observations. The model shows significant improvements in describing the spatial variability (at different spectral ordinates), particularly for the fore-arc area of the Carpathians where a deep sedimentary basin is located. Furthermore, the model presented herein improves estimates of ground shaking at longer spectral ordinates (>1 s) in agreement with the observations. The proposed ground-motion models are valid for hypocentral distances less than 500 km, depths over 70 km and within the moment magnitude range of 4.0–7.4.

Modeling riparian species occurrence from historical surveys to guide restoration planning in northwestern USA

AbstractSuccessful restoration of riparian habitats and functions depends in part on selection of plant species that are suited to local geomorphic and climatic conditions, which often relies on contemporary reference sites to characterize target riparian vegetation communities. In heavily modified landscapes, a lack of undisturbed sites hinders the description of reference conditions to help guide planning efforts. In lieu of contemporary reference sites, we used historical Public Land Survey data from the late 1800s and early 1900s to document historical streamside vegetation at 1685 sites distributed throughout the Columbia River basin. We used those data to construct a random forest classification model using climatic and geomorphic variables to predict the probability of occurrence of riparian vegetation groups (conifer, deciduous, shrub, and willow) and individual taxa (fir, pine, cedar, cottonwood, alder, sagebrush) for all stream reaches with bankfull width >6 m in the interior Columbia River basin. The most common predictor variables included in the best models for vegetation groups or individual taxa were mean annual precipitation, minimum temperature, elevation, and bankfull width. For some taxa, temperature range and floodplain width were also important predictors. The probability maps indicate that riparian zones were likely dominated by willow species in semi‐desert regions and by conifer species in the humid mountain regions. Deciduous species dominated riparian areas in transition zones between conifer forests and semi‐deserts. Species distributions suggest that streams in the semi‐deserts were likely characterized by little shade and low wood abundance, whereas streams in the humid mountains would have been more heavily shaded with high wood abundance. The transitional deciduous areas were likely shaded with moderate wood abundance. Historical trends in air temperature and precipitation suggest relatively small changes in climate since the time of the surveys, indicating that current species ranges are likely similar to historical species ranges. Hence, these maps can be used to help identify suitable taxonomic groups and expected riparian functions for riparian restoration in the Columbia River basin, with appropriate adjustments made to site‐specific restoration designs to account for model uncertainty, future climate change, or land use constraints.

Predicting Communicative Participation in Adults Across Communication Disorders.

Purpose The purpose of this study was to explore the extent to which communicative participation differs across diagnoses and if there are common predictor variables for communicative participation across diagnoses. Method Survey data on self-report variables including communicative participation were collected from 141 community-dwelling adults with communication disorders due to Parkinson's disease, cerebrovascular accident, spasmodic dysphonia, or vocal fold immobility (VFI). Analysis of covariance was used to determine communicative participation differences between diagnoses, with age, sex, and hearing status as covariates. Sequential entry linear regression was used to examine associations between communicative participation and variables representing a range of psychosocial constructs across diagnoses. Results The VFI group had the least favorable communicative participation differing significantly from Parkinson's disease and spasmodic dysphonia groups. Self-rated speech/voice severity, self-rated effort, mental health, perceived social support, and resilience contributed to variance in communicative participation when pooled across diagnoses. The relationship between communicative participation and the variables of effort and resilience differed significantly when diagnosis was considered. Conclusions The findings suggest that communicative participation restrictions may vary across some diagnoses but not others. People with VFI appear to differ from other diagnosis groups in the extent of participation restrictions. Effort and resilience may play different roles in contributing to communicative participation in different disorders, but constructs such as social support, severity, and mental health appear to have consistent relationships with communicative participation across diagnoses. The findings can help clinicians identify psychosocial factors beyond the impairment that impact clients' communication in daily situations.

Sugar Maple, Red Maple, and Yellow Birch Growth and Mortality in Even-Aged Adirondack Northern Hardwoods

AbstractTree diameter growth models for northern hardwoods commonly used large data sets representing a composite of stands with varying management histories, structural characteristics, and age distributions. Yet common predictor variables like diameter can show differences in growth patterns for stands with different age structures and management histories. To address that, we modeled growth and mortality for sugar maple, red maple, and yellow birch in thinned even-aged Adirondack northern hardwoods. Findings indicate that change in diameter depends on initial diameter for sugar maple, with the rate decreasing exponentially from the largest size class to the smallest. Initial diameter did not prove significant with red maple and yellow birch in these thinned stands, perhaps because of the limited sample of trees of small diameter. Stand relative density and time since treatment affected growth for all three species. Those variables also proved significant for predicting mortality of sugar maple. Analyses revealed fewer losses of sugar and red maples among the larger diameter classes, but no relationship with diameter for yellow birch. Plot relative density did not affect mortality with red maple, but time after thinning had a significant effect on survival of all species.

Modeling the risk factors for dyslipidemia and blood lipid indices: Ravansar cohort study

BackgroundLipid disorder is one of the most important risk factors for chronic diseases. Identifying the factors affecting the development of lipid disorders helps reduce chronic diseases, especially Chronic Heart Disease (CHD). The aim of this study was to model the risk factors for dyslipidemia and blood lipid indices.MethodsThis study was conducted based on the data collected in the initial phase of Ravansar cohort study (2014–16). At the beginning, all the 453 available variables were examined in 33 stages of sensitivity analysis by perceptron Artificial Neural Network (ANN) data mining model. In each stage, the variables that were more important in the diagnosis of dyslipidemia were identified. The relationship among the variables was investigated using stepwise regression. The data obtained were analyzed in SPSS software version 25, at 0.05 level of significance.ResultsForty percent of the subjects were diagnosed with lipid disorder. ANN identified 12 predictor variables for dyslipidemia related to nutrition and physical status. Alkaline phosphatase, Fat Free Mass (FFM) index, and Hemoglobin (HGB) had a significant relationship with all the seven blood lipid markers. The Waist Hip Ratio was the most effective variable that showed a stronger correlation with cholesterol and Low-Density Lipid (LDL). The FFM index had the greatest effect on triglyceride, High-Density Lipid (HDL), cholesterol/HDL, triglyceride/HDL, and LDL/HDL. The greatest coefficients of determination pertained to the triglyceride/HDL (0.203) and cholesterol/HDL (0.188) model with nine variables and the LDL/HDL (0.180) model with eight variables.ConclusionAccording to the results, alkaline phosphatase, FFM index, and HGB were three common predictor variables for all the blood lipid markers. Specialists should focus on controlling these factors in order to gain greater control over blood lipid markers.

Exploring health-related quality of life of Spanish domestic adult adoptees: sociodemographic characteristics, chronic medical conditions, and gender differences.

The main objective is to study the health-related quality of life (HRQoL) as a positive approach to health in adult adoptees. We will also consider comparison with population norms, relation to chronic medical conditions, the role of sociodemographic variables, and gender differences among the variables in this study. This was a cross-sectional study of 179 adult domestic adoptees from Spain aged between 18 and 44. They were recruited from a longitudinal study, which was on its third wave. They answered an interview and the self-report measure Health Survey Short Form (SF-12). ANOVAs and t test analyses to compare groups, and multiple regression to determine predictors of HRQoL, were performed. Adoptees are not different from the general population in terms of perceived physical health (PH) and mental health (MH). Both PH and MH are related to different variables (e.g., current age is negatively associated with PH, as is the number of chronic medical conditions with MH). Despite the absence of gender differences in MH perception, there were common and different predictor variables for males and females. The number of chronic medical conditions predicts HRQoL regardless of gender. For women, employment is also a predictor, and for men, other variables are being in a romantic relationship, having children, and low income. Adult adoptees are comparable to general population in terms of HRQoL, but there are associated variables (chronic medical conditions and gender) that should be considered. Post-adoption services should promote health and support in life transitions for this group.

Development of a land use regression model for daily NO2 and NOx concentrations in the Brisbane metropolitan area, Australia

Land use regression models are an established method for estimating spatial variability in gaseous pollutant levels across urban areas. Existing LUR models have been developed to predict annual average concentrations of airborne pollutants. None of those models have been developed to predict daily average concentrations, which are useful in health studies focused on the acute impacts of air pollution. In this study, we developed LUR models to predict daily NO2 and NOx concentrations during 2009–2012 in the Brisbane Metropolitan Area (BMA), Australia's third-largest city. The final models explained 64% and 70% of spatial variability in NO2 and NOx, respectively, with leave-one-out-cross-validation R2 of 3–49% and 2–51%. Distance to major road and industrial area were the common predictor variables for both NO2 and NOx, suggesting an important role for road traffic and industrial emissions. The novel modeling approach adopted here can be applied in other urban locations in epidemiological studies.

Comparison and Correction of Modulus Reduction Models for Clays and Silts

AbstractPrediction models of normalized modulus reduction curves (G/Gmax) for clays and silts have been proposed by several researchers in the past decades. However, model uncertainties have been recognized in these studies. This study compares five prediction models of G/Gmax from previous studies, which use common predictor variables of cyclic shear strain amplitude, effective confining stress, and plasticity index. The relative differences between these models are described against these predictor variables. Then the model biases are evaluated through residual analyses using the modulus reduction database. Modulus reductions of G are also discussed by combining the prediction models of G/Gmax and Gmax to illustrate the importance of these combinations. Correlations among the residuals of ln(G) are calculated, and the correction factors of the G/Gmax model are presented conditioned on the Gmax and shear strength measurements.

Predicting the Occurrence of Cave-Inhabiting Fauna Based on Features of the Earth Surface Environment.

One of the most challenging fauna to study in situ is the obligate cave fauna because of the difficulty of sampling. Cave-limited species display patchy and restricted distributions, but it is often unclear whether the observed distribution is a sampling artifact or a true restriction in range. Further, the drivers of the distribution could be local environmental conditions, such as cave humidity, or they could be associated with surface features that are surrogates for cave conditions. If surface features can be used to predict the distribution of important cave taxa, then conservation management is more easily obtained. We examined the hypothesis that the presence of major faunal groups of cave obligate species could be predicted based on features of the earth surface. Georeferenced records of cave obligate amphipods, crayfish, fish, isopods, beetles, millipedes, pseudoscorpions, spiders, and springtails within the area of Appalachian Landscape Conservation Cooperative in the eastern United States (Illinois to Virginia and New York to Alabama) were assigned to 20 x 20 km grid cells. Habitat suitability for these faunal groups was modeled using logistic regression with twenty predictor variables within each grid cell, such as percent karst, soil features, temperature, precipitation, and elevation. Models successfully predicted the presence of a group greater than 65% of the time (mean = 88%) for the presence of single grid cell endemics, and for all faunal groups except pseudoscorpions. The most common predictor variables were latitude, percent karst, and the standard deviation of the Topographic Position Index (TPI), a measure of landscape rugosity within each grid cell. The overall success of these models points to a number of important connections between the surface and cave environments, and some of these, especially soil features and topographic variability, suggest new research directions. These models should prove to be useful tools in predicting the presence of species in understudied areas.

Clinical effectiveness of cognitive behavioral therapy for depression in routine care: A propensity score based comparison between randomized controlled trials and clinical practice

BackgroundThe efficacy of cognitive behavioral therapy (CBT) for the treatment of depressive disorders has been demonstrated in many randomized controlled trials (RCTs). This study investigated whether for CBT similar effects can be expected under routine care conditions when the patients are comparable to those examined in RCTs. MethodN=574 CBT patients from an outpatient clinic were stepwise matched to the patients undergoing CBT in the National Institute of Mental Health Treatment of Depression Collaborative Research Program (TDCRP). First, the exclusion criteria of the RCT were applied to the naturalistic sample of the outpatient clinic. Second, propensity score matching (PSM) was used to adjust the remaining naturalistic sample on the basis of baseline covariate distributions. Matched samples were then compared regarding treatment effects using effect sizes, average treatment effect on the treated (ATT) and recovery rates. ResultsCBT in the adjusted naturalistic subsample was as effective as in the RCT. However, treatments lasted significantly longer under routine care conditions. LimitationsThe samples included only a limited amount of common predictor variables and stemmed from different countries. There might be additional covariates, which could potentially further improve the matching between the samples. ConclusionsCBT for depression in clinical practice might be equally effective as manual-based treatments in RCTs when they are applied to comparable patients. The fact that similar effects under routine conditions were reached with more sessions, however, points to the potential to optimize treatments in clinical practice with respect to their efficiency.

Social identity and applicant attraction: Exploring the role of multiple levels of self

SummaryApplicant attraction is a critical objective of recruitment. Common predictor variables of applicant attraction are limited in that they do not provide a comprehensive understanding of the process that shapes the perceptions and beliefs of job applicants about the firms for which they aspire to work for. Because individuals have the inherent desire to expand and enhance their social identities (e.g., personal, relational, and collective identities), they are likely to be attracted to organizations that allow them to do so. Building on recent work on levels of self, our paper suggests that social identities mediate the relation between currently established predictor variables of applicant attraction (e.g., compensation, type of work, and organizational image) and important applicant attraction outcomes. Common predictor variables of applicant attraction can lead to the activation, evaluation, and identification processes described by social identity theory. A theoretical framework is presented that illustrates the mediating influence of social identity on the relations between common predictor variables and applicant attraction outcomes. This framework may lead to more effective recruitment strategies. Copyright © 2015 John Wiley & Sons, Ltd.

Consumption, change in expectations and equity returns

Using quarterly data for the period 1959 to 2008, I study the relationship between excess stock returns and the change in expectations of the consumption–wealth ratio and of future long-run consumption growth. Using a vector error correction model (VECM), I estimate revisions in expectations on the consumption–wealth ratio and on the discounted value of future consumption growth; the latter being of high relevance in the asset pricing literature but difficult to identify empirically. My findings show that these variables are strong predictors of future excess stock returns when compared to several common predictor variables. Furthermore, these results seem to be robust in out-of-sample and in-sample analyses, and appear not to be driven by persistence.

Developing and evaluating land use regression models for NO2 in city-wide, regional, and continental scales in Europe

Background: The applications of land use regression (LUR) models were mainly focused on the intra-urban level. However, developing models based on several areas with geographical similarity and consistency of the input data may explain the distributions of air pollution better due to the inclusion of a larger number of sampling sites. Aim: We aim to explore the feasibility of developing LUR models in larger areas and to compare the model performances systematically in different geographical scales in Europe. Methods: We developed LUR models for NO2 at regional and continental scales based on 23 ESCAPE (European Study of Cohorts for Air Pollution Effects) study areas with pooled data and standard protocols. Models in different spatial scales were evaluated with cross validation (CV) and hold-out validation (HV) and were compared with the performances of the city/country-specific models. In addition, we investigated the prediction abilities of the models in independent areas from where no measurements were used for model building. Results: The overall model performances in terms of the model R2 and CV R2 were highest for models at local scale (model :0.85, CV: 0.79) and lowest for models at continental scale (model: 0.57, CV: 0.55). However, the regional scale models (HV:0.39~0.69) revealed higher predictive power in the independent sites than the local models (HV:0.32~0.59) except for the regional model in south Europe where the local models performed better (HV local: 0.64; region: 0.49). The transferable models explained a reasonable fraction of variability in independent areas (HV region: 0.48; EU: 0.53). Conclusion: It is possible to develop LUR models combining measurements from several study areas which outperform the city/country specific models in terms of the prediction ability to the external validation sites. These models can also be transferred to predict the NO2 concentrations in cities with similar geographical conditions and common predictor variables where no measurement data are available.