The objective of this study was to compare performance of logistic regression (LR) with machine learning (ML) for clinical prediction modeling in the literature.
We conducted a Medline literature ...search (1/2016 to 8/2017) and extracted comparisons between LR and ML models for binary outcomes.
We included 71 of 927 studies. The median sample size was 1,250 (range 72–3,994,872), with 19 predictors considered (range 5–563) and eight events per predictor (range 0.3–6,697). The most common ML methods were classification trees, random forests, artificial neural networks, and support vector machines. In 48 (68%) studies, we observed potential bias in the validation procedures. Sixty-four (90%) studies used the area under the receiver operating characteristic curve (AUC) to assess discrimination. Calibration was not addressed in 56 (79%) studies. We identified 282 comparisons between an LR and ML model (AUC range, 0.52–0.99). For 145 comparisons at low risk of bias, the difference in logit(AUC) between LR and ML was 0.00 (95% confidence interval, −0.18 to 0.18). For 137 comparisons at high risk of bias, logit(AUC) was 0.34 (0.20–0.47) higher for ML.
We found no evidence of superior performance of ML over LR. Improvements in methodology and reporting are needed for studies that compare modeling algorithms.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UILJ, UL, UM, UPCLJ, UPUK, ZAGLJ, ZRSKP
Logistic regression studies which assess landslide susceptibility are widely available in the literature. However, a global review of these studies to synthesise and compare the results does not ...exist. There are currently no guidelines for the selection of covariates to be used in logistic regression analysis, and as such, the covariates selected vary widely between studies. An inventory of significant covariates associated with landsliding produced from the full set of such studies globally would be a useful aid to the selection of covariates in future logistic regression studies. Thus, studies using logistic regression for landslide susceptibility estimation published in the literature were collated, and a database was created of the significant factors affecting the generation of landslides. The database records the paper the data were taken from, the year of publication, the approximate longitude and latitude of the study area, the trigger method (where appropriate) and the most dominant type of landslides occurring in the study area. The significant and non-significant (at the 95 % confidence level) covariates were recorded, as well as their coefficient, statistical significance and unit of measurement. The most common statistically significant covariate used in landslide logistic regression was slope, followed by aspect. The significant covariates related to landsliding varied for earthquake-induced landslides compared to rainfall-induced landslides, and between landslide type. More importantly, the full range of covariates used was identified along with their frequencies of inclusion. The analysis showed that there needs to be more clarity and consistency in the methodology for selecting covariates for logistic regression analysis and in the metrics included when presenting the results. Several recommendations for future studies were given.
Urban mobility research is crucial due to the complex and dynamic nature of cities. Traditional zoning methods often overlook the temporal dynamics of urban space usage. This paper aims to ...investigate the daily rhythm of urban space usage and how built environmental features affect spatiotemporal mobility. The methodology integrates various urban data sources, including street and building vectors and points of interest (POI). In Ankara, Türkiye, temporal usage types were identified using principle component analysis and k-means algorithms via Google Maps location-based service data. Multinomial logistic regression model was used to examine the impact of built-environment factors on daily mobility patterns. Results reveal distinct daily usage rhythms, categorizing busy areas into morning, noon, early evening, late evening, diurnal, and nocturnal zones. The study finds: (1) temporal usage is closely associated with spatial characteristics and resident income; (2) work-related activities drive morning mobility, while nighttime mobility relates to residence, income, and accessibility; and (3) population density alone does not guarantee continuous activities; diverse POIs alongside high density are crucial. The identified relationship between mobility measures and urban built environment indicators provides a comprehensive understanding of spatiotemporal variations, offering insights for critical policy evaluations and proposals for the future of urban areas.
•We use Google Maps' mobile phone location request records to uncover the spatiotemporal dynamics of urban space usage.•The impact of the built environment on human temporal activity is analyzed using multinomial logistic regression.•Human mobility areas have different daily rhythms of urban space usage, with different temporal evolutions.•Temporal evolution factors include spatial function and residents' income.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UILJ, UL, UM, UPCLJ, UPUK, ZAGLJ, ZRSKP
While renewable energy technologies (RET) increase their share in power generation systems worldwide, some questions remain open, namely those concerning the opinion of the populations on new ...projects of these technologies. Given the long period of planning and large capital sums required by RET and, in some cases, the fact of being subsidized, it is desirable for decision-makers to acknowledge the public opinion and at least perceive if the opinions are rooted on biased perceptions. In this paper we propose a methodology for public perception and awareness assessment, involving an initial phase of data collection by means of a survey, followed by a phase of regression models construction resulting in predictive models of expected perceptions and attitudes towards RET. The models were translated in a free and easy to use computational Excel application and its usefulness was demonstrated for the case of four electricity RET in Portugal: hydro, wind, biomass and solar.
•Models for expected perception and attitudes towards renewable energy are proposed.•Ordered logistic and binary logistic regression models were used.•A survey for the Portuguese system was used to test the models implementation.•Models are demonstrated to be useful for hydro, wind, sun and biomass power.•Model results are useful for energy policy decision and investment decision making.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UL, UM, UPCLJ, UPUK, ZRSKP
This study posits that all innovations meet consumer resistance, and overcoming this opposition must occur prior to product adoption. Factors driving service innovation resistance remain unclear. To ...better understand this behavior, the present study examines how five theory-driven adoption barriers—usage, value, risk, tradition, and image – as well as three consumer demographics—gender, age, and income—influence consumer adoption versus rejection decisions in Internet and mobile banking. Data from two large nationwide surveys conducted in Finland (n=1736 consumers) test hypotheses using binary logit models comparing mobile banking adopters versus non-adopters, mobile banking postponers versus rejecters, and Internet banking postponers versus rejecters. Study results find that the value barrier is the strongest inhibitor of Internet and mobile banking adoption. In addition, the image barrier slows mobile banking adoption, and the tradition barrier explains the rejection of Internet banking. Gender and age significantly predict adoption and rejection decisions. The results demonstrate notable differences between these seemingly similar service innovations.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UL, UM, UPCLJ, UPUK, ZRSKP
Previous research has established that higher levels of trait Honesty-Humility (HH) are associated with less dishonest behavior in cheating paradigms. However, only imprecise effect size estimates of ...this HH-cheating link are available. Moreover, evidence is inconclusive on whether other basic personality traits from the HEXACO or Big Five models are associated with unethical decision making and whether such effects have incremental validity beyond HH. We address these issues in a highly powered reanalysis of 16 studies assessing dishonest behavior in an incentivized, one-shot cheating paradigm (N = 5,002). For this purpose, we rely on a newly developed logistic regression approach for the analysis of nested data in cheating paradigms. We also test theoretically derived interactions of HH with other basic personality traits (i.e., Emotionality and Conscientiousness) and situational factors (i.e., the baseline probability of observing a favorable outcome) as well as the incremental validity of HH over demographic characteristics. The results show a medium to large effect of HH (odds ratio = 0.53), which was independent of other personality, situational, or demographic variables. Only one other trait (Big Five Agreeableness) was associated with unethical decision making, although it failed to show any incremental validity beyond HH.
Full text
Available for:
CEKLJ, DOBA, IZUM, KILJ, NUK, PILJ, PNG, SAZU, UILJ, UKNU, UL, UM, UPUK
Multiple Criteria Evaluation (MCE) is a multi-attributes decision making tool often used in land suitability analysis and land use simulation using Cellular Automata (CA)-Markov model. The goal of ...this research is to explore the feasibility of using historical data of a study area to select, score, and weight factors quantitatively in the MCE. We have developed logistic regression models fitted by the historical land use changes to select and score each potential factor, and used the Entropy method to determine weights for the selected factors. The MCE output is then used as the input of CA-Markov model to simulate land use changes from 2001 to 2011. The land use simulation result was compared against observed 2011 land use in order to examine the performance of the updated MCE method. The result shows that the use of MCE factors derived from historical data produces reasonable goodness of fit, based on current literature. The major advantage of the updated MCE method is that the factor selection, scores, and weights are all derived from local data reflecting the actual historical trend. This quantitative approach also allows one to efficiently calibrate CA-Markov model and develop different land use planning scenarios by adjusting scores and weights for different factors with the knowledge of historical change.
•Proposed a logistic regression model to quantitatively analyze historical data for selecting suitability factors and assigning location-specific suitability scores in MCE.•Used the Entropy method to objectively determine the weights for selected factors.•Validated the performance of the updated MCE by CA-Markov model and produced reasonable goodness of fit.•Discussed the further applicability on CA-Markov model calibration and land use scenario planning.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UL, UM, UPCLJ, UPUK, ZRSKP
The Random Forest (RF) algorithm for regression and classification has considerably gained popularity since its introduction in 2001. Meanwhile, it has grown to a standard classification approach ...competing with logistic regression in many innovation-friendly scientific fields.
In this context, we present a large scale benchmarking experiment based on 243 real datasets comparing the prediction performance of the original version of RF with default parameters and LR as binary classification tools. Most importantly, the design of our benchmark experiment is inspired from clinical trial methodology, thus avoiding common pitfalls and major sources of biases.
RF performed better than LR according to the considered accuracy measured in approximately 69% of the datasets. The mean difference between RF and LR was 0.029 (95%-CI =0.022,0.038) for the accuracy, 0.041 (95%-CI =0.031,0.053) for the Area Under the Curve, and - 0.027 (95%-CI =-0.034,-0.021) for the Brier score, all measures thus suggesting a significantly better performance of RF. As a side-result of our benchmarking experiment, we observed that the results were noticeably dependent on the inclusion criteria used to select the example datasets, thus emphasizing the importance of clear statements regarding this dataset selection process. We also stress that neutral studies similar to ours, based on a high number of datasets and carefully designed, will be necessary in the future to evaluate further variants, implementations or parameters of random forests which may yield improved accuracy compared to the original version with default values.
Full text
Available for:
DOBA, IZUM, KILJ, NUK, PILJ, PNG, SAZU, SIK, UILJ, UKNU, UL, UM, UPUK
Different study designs and population size may require different sample size for logistic regression. This study aims to propose sample size guidelines for logistic regression based on observational ...studies with large population.
We estimated the minimum sample size required based on evaluation from real clinical data to evaluate the accuracy between statistics derived and the actual parameters. Nagelkerke r-squared and coefficients derived were compared with their respective parameters.
With a minimum sample size of 500, results showed that the differences between the sample estimates and the population was sufficiently small. Based on an audit from a medium size of population, the differences were within ± 0.5 for coefficients and ± 0.02 for Nagelkerke
-squared. Meanwhile for large population, the differences are within ± 1.0 for coefficients and ± 0.02 for Nagelkerke
-squared.
For observational studies with large population size that involve logistic regression in the analysis, taking a minimum sample size of 500 is necessary to derive the statistics that represent the parameters. The other recommended rules of thumb are EPV of 50 and formula;
= 100 + 50
where
refers to number of independent variables in the final model.
Full text
Available for:
FFLJ, IZUM, KILJ, NUK, ODKLJ, PILJ, PNG, SAZU, UL, UM, UPUK
•Conducted ecological momentary assessment with residents of four European cities.•Positive mood improved within 10 min of natural environment exposure.•Findings varied by city, gender, age and ...residential green space exposure.•Negative mood declined within 10 min of natural environment exposure.•Weaker associations were found within 30 min of natural environment exposure.
Exposure to natural outdoor environments (NOE) has been shown in population-level studies to reduce anxiety and psychological distress. This study investigated how exposure to one’s everyday natural outdoor environments over one week influenced mood among residents of four European cities including Barcelona (Spain), Stoke-on-Trent (United Kingdom), Doetinchem (The Netherlands) and Kaunas (Lithuania). Participants (n = 368) wore a smartphone equipped with software applications to track location and mood (using mobile ecological momentary assessment (EMA) software), for seven consecutive days. We estimated random-effects ordered logistic regression models to examine the association between mood (positive and negative affect), and exposure to green space, represented by two binary variables indicating exposure versus no exposure to NOE using GPS tracking and satellite and aerial imagery, 10 and 30 min prior to participants’ completing the EMA. Models were adjusted for home city, day of the week, hour of the day, EMA survey type, residential NOE exposure, and sex, age, education level, mental health status and neighbourhood socioeconomic status. In addition, we tested for heterogeneity of effect by city, sex, age, residential NOE exposure and mental health status. Within 10 min of NOE exposure, compared to non-exposure, we found that overall there was a positive relationship with positive affect (OR: 1.39, 95% CI: 1.06, 1.81) of EMA surveys, and non-significant negative association with negative affect (OR: 0.80, 95% CI: 0.58, 1.10). When stratifying, associations were consistently found for Stoke-on-Trent inhabitants and men, while findings by age group were inconsistent. Weaker and less consistent associations were found for exposure 30 min prior to EMA. Our findings support increasing evidence of psychological and mental health benefits of exposure to natural outdoor environments, especially among urban populations such as those included in our study.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UILJ, UL, UM, UPCLJ, UPUK, ZAGLJ, ZRSKP