A comparison of model choice strategies for logistic regression

Purpose:The purpose of this study is to develop and compare model choice strategies in context of logistic regression.Model choice means the choice of the covariates to be included in the model.Design/methodology/approach:The study is based on Monte Carlo simulations.The methods are compared in terms of three measures of accuracy:specificity and two kinds of sensitivity.A loss function combining sensitivity and specificity is introduced and used for a final comparison.Findings:The choice of method depends on how much the users emphasize sensitivity against specificity.It also depends on the sample size.For a typical logistic regression setting with a moderate sample size and a small to moderate effect size,either BIC,BICc or Lasso seems to be optimal.Research limitations:Numerical simulations cannot cover the whole range of data-generating processes occurring with real-world data.Thus,more simulations are needed.Practical implications:Researchers can refer to these results if they believe that their data-generating process is somewhat similar to some of the scenarios presented in this paper.Alternatively,they could run their own simulations and calculate the loss function.Originality/value:This is a systematic comparison of model choice algorithms and heuristics in context of logistic regression.The distinction between two types of sensitivity and a comparison based on a loss function are methodological novelties.

Country-based modelling of COVID-19 case fatality rate:A multiple regression analysis

BACKGROUND The spread of the severe acute respiratory syndrome coronavirus 2 outbreak worldwide has caused concern regarding the mortality rate caused by the infection.The determinants of mortality on a global scale cannot be fully understood due to lack of information.AIM To identify key factors that may explain the variability in case lethality across countries.METHODS We identified 21 Potential risk factors for coronavirus disease 2019(COVID-19)case fatality rate for all the countries with available data.We examined univariate relationships of each variable with case fatality rate(CFR),and all independent variables to identify candidate variables for our final multiple model.Multiple regression analysis technique was used to assess the strength of relationship.RESULTS The mean of COVID-19 mortality was 1.52±1.72%.There was a statistically significant inverse correlation between health expenditure,and number of computed tomography scanners per 1 million with CFR,and significant direct correlation was found between literacy,and air pollution with CFR.This final model can predict approximately 97%of the changes in CFR.CONCLUSION The current study recommends some new predictors explaining affect mortality rate.Thus,it could help decision-makers develop health policies to fight COVID-19.

A Hybrid Model Evaluation Based on PCA Regression Schemes Applied to Seasonal Precipitation Forecast

Possible changes in the structure and seasonal variability of the subtropical ridge may lead to changes in the rainfall’s variability modes over Caribbean region. This generates additional difficulties around water resource planning, therefore, obtaining seasonal prediction models that allow these variations to be characterized in detail, it’s a concern, specially for island states. This research proposes the construction of statistical-dynamic models based on PCA regression methods. It is used as predictand the monthly precipitation accumulated, while the predictors (6) are extracted from the ECMWF-SEAS5 ensemble mean forecasts with a lag of one month with respect to the target month. In the construction of the models, two sequential training schemes are evaluated, obtaining that only the shorter preserves the seasonal characteristics of the predictand. The evaluation metrics used, where cell-point and dichotomous methodologies are combined, suggest that the predictors related to sea surface temperatures do not adequately represent the seasonal variability of the predictand, however, others such as the temperature at 850 hPa and the Outgoing Longwave Radiation are represented with a good approximation regardless of the model chosen. In this sense, the models built with the nearest neighbor methodology were the most efficient. Using the individual models with the best results, an ensemble is built that allows improving the individual skill of the models selected as members by correcting the underestimation of precipitation in the dynamic model during the wet season, although problems of overestimation persist for thresholds lower than 50 mm.

Modeling of Total Dissolved Solids (TDS) and Sodium Absorption Ratio (SAR) in the Edwards-Trinity Plateau and Ogallala Aquifers in the Midland-Odessa Region Using Random Forest Regression and eXtreme Gradient Boosting

Efficient water quality monitoring and ensuring the safety of drinking water by government agencies in areas where the resource is constantly depleted due to anthropogenic or natural factors cannot be overemphasized. The above statement holds for West Texas, Midland, and Odessa Precisely. Two machine learning regression algorithms (Random Forest and XGBoost) were employed to develop models for the prediction of total dissolved solids (TDS) and sodium absorption ratio (SAR) for efficient water quality monitoring of two vital aquifers: Edward-Trinity (plateau), and Ogallala aquifers. These two aquifers have contributed immensely to providing water for different uses ranging from domestic, agricultural, industrial, etc. The data was obtained from the Texas Water Development Board (TWDB). The XGBoost and Random Forest models used in this study gave an accurate prediction of observed data (TDS and SAR) for both the Edward-Trinity (plateau) and Ogallala aquifers with the R2 values consistently greater than 0.83. The Random Forest model gave a better prediction of TDS and SAR concentration with an average R, MAE, RMSE and MSE of 0.977, 0.015, 0.029 and 0.00, respectively. For the XGBoost, an average R, MAE, RMSE, and MSE of 0.953, 0.016, 0.037 and 0.00, respectively, were achieved. The overall performance of the models produced was impressive. From this study, we can clearly understand that Random Forest and XGBoost are appropriate for water quality prediction and monitoring in an area of high hydrocarbon activities like Midland and Odessa and West Texas at large.

Utilization of Logistical Regression to the Modified Sine-Gordon Model in the MST Experiment

In this paper, a logistical regression statistical analysis (LR) is presented for a set of variables used in experimental measurements in reversed field pinch (RFP) machines, commonly known as “slinky mode” (SM), observed to travel around the torus in Madison Symmetric Torus (MST). The LR analysis is used to utilize the modified Sine-Gordon dynamic equation model to predict with high confidence whether the slinky mode will lock or not lock when compared to the experimentally measured motion of the slinky mode. It is observed that under certain conditions, the slinky mode “locks” at or near the intersection of poloidal and/or toroidal gaps in MST. However, locked mode cease to travel around the torus;while unlocked mode keeps traveling without a change in the energy, making it hard to determine an exact set of conditions to predict locking/unlocking behaviour. The significant key model parameters determined by LR analysis are shown to improve the Sine-Gordon model’s ability to determine the locking/unlocking of magnetohydrodyamic (MHD) modes. The LR analysis of measured variables provides high confidence in anticipating locking versus unlocking of slinky mode proven by relational comparisons between simulations and the experimentally measured motion of the slinky mode in MST.

Research on the Relationship Between Average Cigarette Price per Box and Government Procurement in City A Based on a Regression Model

This study aims to analyze and predict the relationship between the average price per box in the cigarette market of City A and government procurement,providing a scientific basis and support for decision-making.By reviewing relevant theories and literature,qualitative prediction methods,regression prediction models,and other related theories were explored.Through the analysis of annual cigarette sales data and government procurement data in City A,a comprehensive understanding of the development of the tobacco industry and the economic trends of tobacco companies in the county was obtained.By predicting and analyzing the average price per box of cigarette sales across different years,corresponding prediction results were derived and compared with actual sales data.The prediction results indicate that the correlation coefficient between the average price per box of cigarette sales and government procurement is 0.982,implying that government procurement accounts for 96.4%of the changes in the average price per box of cigarettes.These findings offer an in-depth exploration of the relationship between the average price per box of cigarettes in City A and government procurement,providing a scientific foundation for corporate decision-making and market operations.

Genetic Regression Model for Dam Safety Monitoring

Under-fitting problems usually occur in regression models for dam safety monitoring.To overcome the local convergence of the regression, a genetic algorithm (GA) was proposed using a real parameter coding, a ranking selection operator, an arithmetical crossover operator and a uniform mutation operator, and calculated the least-square error of the observed and computed values as its fitness function. The elitist strategy was used to improve the speed of the convergence. After that, the modified genetic algorithm was applied to reassess the coefficients of the regression model and a genetic regression model was set up. As an example, a slotted gravity dam in the Northeast of China was introduced. The computational results show that the genetic regression model can solve the under-fitting problems perfectly.

RBF neural network regression model based on fuzzy observations

A fuzzy observations-based radial basis function neural network （FORBFNN） is presented for modeling nonlinear systems in which the observations of response are imprecise but can be represented as fuzzy membership functions. In the FORBFNN model, the weight coefficients of nodes in the hidden layer are identified by using the fuzzy expectation-maximization （ EM ） algorithm, whereas the optimal number of these nodes as well as the centers and widths of radial basis functions are automatically constructed by using a data-driven method. Namely, the method starts with an initial node, and then a new node is added in a hidden layer according to some rules. This procedure is not terminated until the model meets the preset requirements. The method considers both the accuracy and complexity of the model. Numerical simulation results show that the modeling method is effective, and the established model has high prediction accuracy.

Quality prediction of batch process using the global-local discriminant analysis based Gaussian process regression model

The conventional single model strategy may be ill- suited due to the multiplicity of operation phases and system uncertainty. A novel global-local discriminant analysis （GLDA） based Gaussian process regression （GPR） approach is developed for the quality prediction of nonlinear and multiphase batch processes. After the collected data is preprocessed through batchwise unfolding, the hidden Markov model （HMM） is applied to identify different operation phases. A GLDA algorithm is also presented to extract the appropriate process variables highly correlated with the quality variables, decreasing the complexity of modeling. Besides, the multiple local GPR models are built in the reduced- dimensional space for all the identified operation phases. Furthermore, the HMM-based state estimation is used to classify each measurement sample of a test batch into a corresponding phase with the maximal likelihood estimation. Therefore, the local GPR model with respect to specific phase is selected for online prediction. The effectiveness of the proposed prediction approach is demonstrated through the multiphase penicillin fermentation process. The comparison results show that the proposed GLDA-GPR approach is superior to the regular GPR model and the GPR based on HMM （HMM-GPR） model.

Selection of the Linear Regression Model According to the Parameter Estimation

In this paper, based on the theory of parameter estimation, we give a selection method and, in a sense of a good character of the parameter estimation, we think that it is very reasonable. Moreover, we offer a calculation method of selection statistic and an applied example.

Combined model based on optimized multi-variable grey model and multiple linear regression

The construction method of background value is improved in the original multi-variable grey model （MGM（1,m）） from its source of construction errors. The MGM（1,m） with optimized background value is used to eliminate the random fluctuations or errors of the observational data of all variables, and the combined prediction model together with the multiple linear regression is established in order to improve the simulation and prediction accuracy of the combined model. Finally, a combined model of the MGM（1,2） with optimized background value and the binary linear regression is constructed by an example. The results show that the model has good effects for simulation and prediction.

Small-time scale network traffic prediction based on a local support vector machine regression model

In this paper we apply the nonlinear time series analysis method to small-time scale traffic measurement data. The prediction-based method is used to determine the embedding dimension of the traffic data. Based on the reconstructed phase space, the local support vector machine prediction method is used to predict the traffic measurement data, and the BIC-based neighbouring point selection method is used to choose the number of the nearest neighbouring points for the local support vector machine regression model. The experimental results show that the local support vector machine prediction method whose neighbouring points are optimized can effectively predict the small-time scale traffic measurement data and can reproduce the statistical features of real traffic measurements.

Semiparametric Regression and Model Refining

This paper presents a semiparametric adjustment method suitable for general cases.Assuming that the regularizer matrix is positive definite,the calculation method is discussed and the corresponding formulae are presented.Finally,a simulated adjustment problem is constructed to explain the method given in this paper.The results from the semiparametric model and G_M model are compared.The results demonstrate that the model errors or the systematic errors of the observations can be detected correctly with the semiparametric estimate method.

Landslide susceptibility mapping using an integrated model of information value method and logistic regression in the Bailongjiang watershed,Gansu Province,China

Bailongjiang watershed in southern Gansu province, China, is one of the most landslide-prone regions in China, characterized by very high frequency of landslide occurrence. In order to predict the landslide occurrence, a comprehensive map of landslide susceptibility is required which may be significantly helpful in reducing loss of property and human life. In this study, an integrated model of information value method and logistic regression is proposed by using their merits at maximum and overcoming their weaknesses, which may enhance precision and accuracy of landslide susceptibility assessment. A detailed and reliable landslide inventory with 1587 landslides was prepared and randomly divided into two groups,(i) training dataset and(ii) testing dataset. Eight distinct landslide conditioning factors including lithology, slope gradient, aspect, elevation, distance to drainages,distance to faults, distance to roads and vegetation coverage were selected for landslide susceptibility mapping. The produced landslide susceptibility maps were validated by the success rate and prediction rate curves. The validation results show that the success rate and the prediction rate of the integrated model are 81.7 % and 84.6 %, respectively, which indicate that the proposed integrated method is reliable to produce an accurate landslide susceptibility map and the results may be used for landslides management and mitigation.

FUNCTIONAL-COEFFICIENT REGRESSION MODEL AND ITS ESTIMATION

In this paper,a class of functional-coefficient regression models is proposed and an estimation procedure based on the locally weighted least equares is suggested.This class of models,with the proposed estimation method,is a powerful means for exploratory data analysis.

WAVELET ESTIMATION FOR JUMPS IN A HETEROSCEDASTIC REGRESSION MODEL

Wavelets are applied to detect the jumps in a heteroscedastic regression model. It is shown that the wavelet coefficients of the data have significantly large absolute values across fine scale levels near the jump points. Then a procedure is developed to estimate the jumps and jump heights. All estimators are proved to be consistent.

ON CONFIDENCE REGIONS OF SEMIPARAMETRIC NONLINEAR REGRESSION MODELS(A GEOMETRIC APPROACH)

A geometric framework is proposed for semiparametric nonlinear regression models based on the concept of least favorable curve, introduced by Severini and Wong (1992). The authors use this framework to drive three kinds of improved approximate confidence regions for the parameter and parameter subset in terms of curvatures. The results obtained by Hamilton et al. (1982), Hamilton (1986) and Wei (1994) are extended to semiparametric nonlinear regression models.

Comparison Between Radial Basis Function Neural Network and Regression Model for Estimation of Rice Biophysical Parameters Using Remote Sensing

The radial basis function （RBF） emerged as a variant of artificial neural network. Generalized regression neural network （GRNN） is one type of RBF, and its principal advantages are that it can quickly learn and rapidly converge to the optimal regression surface with large number of data sets. Hyperspectral reflectance （350 to 2500 nm） data were recorded at two different rice sites in two experiment fields with two cultivars, three nitrogen treatments and one plant density （45 plants m^-2）. Stepwise multivariable regression model （SMR） and RBF were used to compare their predictability for the leaf area index （LAI） and green leaf chlorophyll density （GLCD） of rice based on reflectance （R） and its three different transformations, the first derivative reflectance （D1）, the second derivative reflectance （D2） and the log-transformed reflectance （LOG）. GRNN based on D1 was the best model for the prediction of rice LAI and CLCD. The relationships between different transformations of reflectance and rice parameters could be further improved when RBF was employed. Owing to its strong capacity for nonlinear mapping and good robustness, GRNN could maximize the sensitivity to chlorophyll content using D1. It is concluded that RBF may provide a useful exploratory and predictive tool for the estimation of rice biophysical parameters.

Evaluation of Inference Adequacy in Cumulative Logistic Regression Models:An Empirical Validation of ISW-Ridge Relationships

Internal solitary wave propagation over a submarine ridge results in energy dissipation, in which the hydrodynamic interaction between a wave and ridge affects marine environment. This study analyzes the effects of ridge height and potential energy during wave-ridge interaction with a binary and cumulative logistic regression model. In testing the Global Null Hypothesis, all values are p 〈0.001, with three statistical methods, such as Likelihood Ratio, Score, and Wald. While comparing with two kinds of models, tests values obtained by cumulative logistic regression models are better than those by binary logistic regression models. Although this study employed cumulative logistic regression model, three probability functions p^1, p^2 and p^3, are utilized for investigating the weighted influence of factors on wave reflection. Deviance and Pearson tests are applied to cheek the goodness-of-fit of the proposed model. The analytical results demonstrated that both ridge height （X1 ） and potential energy （X2 ） significantly impact （p 〈 0. 0001 ） the amplitude-based refleeted rate; the P-values for the deviance and Pearson are all 〉 0.05 （0.2839, 0.3438, respectively）. That is, the goodness-of-fit between ridge height （ X1 ） and potential energy （X2） can further predict parameters under the scenario of the best parsimonious model. Investigation of 6 predictive powers （ R2, Max-rescaled R^2, Sorners＇ D, Gamma, Tau-a, and c, respectively） indicate that these predictive estimates of the proposed model have better predictive ability than ridge height alone, and are very similar to the interaction of ridge height and potential energy. It can be concluded that the goodness-of-fit and prediction ability of the cumulative logistic regression model are better than that of the binary logistic regression model.

Predicting the growth performance of growing-finishing pigs based on net energy and digestible lysine intake using multiple regression and artificial neural networks models

Backgrounds:Evaluating the growth performance of pigs in real-time is laborious and expensive,thus mathematical models based on easily accessible variables are developed.Multiple regression(MR)is the most widely used tool to build prediction models in swine nutrition,while the artificial neural networks(ANN)model is reported to be more accurate than MR model in prediction performance.Therefore,the potential of ANN models in predicting the growth performance of pigs was evaluated and compared with MR models in this study.Results:Body weight(BW),net energy(NE)intake,standardized ileal digestible lysine(SID Lys)intake,and their quadratic terms were selected as input variables to predict ADG and F/G among 10 candidate variables.In the training phase,MR models showed high accuracy in both ADG and F/G prediction(R^(2)_(ADG)=0.929,R^(2)_(F/G)=0.886)while ANN models with 4,6 neurons and radial basis activation function yielded the best performance in ADG and F/G prediction(R^(2)_(ADG)=0.964,R^(2)_(F/G)=0.932).In the testing phase,these ANN models showed better accuracy in ADG prediction(CCC:0.976 vs.0.861,R^(2):0.951 vs.0.584),and F/G prediction(CCC:0.952 vs.0.900,R^(2):0.905 vs.0.821)compared with the MR models.Meanwhile,the“over-fitting”occurred in MR models but not in ANN models.On validation data from the animal trial,ANN models exhibited superiority over MR models in both ADG and F/G prediction(P<0.01).Moreover,the growth stages have a significant effect on the prediction accuracy of the models.Conclusion:Body weight,NE intake and SID Lys intake can be used as input variables to predict the growth performance of growing-finishing pigs,with trained ANN models are more flexible and accurate than MR models.Therefore,it is promising to use ANN models in related swine nutrition studies in the future.