Classical Logistic Regression Model Research Articles

İkili yanıt değişkenine sahip modellerin yeterliliklerine ilişkin benzetim çalışması – parametrik olmayan yöntemler

Regression models used to explore the importance of several explanatory variables in estimation, classification and analytical tools play an efficient role for many data analysis. Although the classical linear model is quite easy to use, it is often not sufficient for many real data sets as the relationships between variables do not hold the assumption of the linearity of the relationship between dependent and explanatory variables. Under this study, a nonparametric model fitting that does not require to form a strict mathematical relationship between dependent and explanatory variables will be discussed on the contrary the assumption in multiple linear regression. In this study, the relationship between a binary dependent variable and the explanatory variables will be examined in a conducted simulation study by using generalized linear, the additive logistic regression in case of classical logistic regression model and decision trees to explore the cause and effect relationship. The methods in question and the simulation study will be performed for small, medium and large data sets when multicollinearity problem exists and will be compared with each other.

Detecting corporate tax evasion using a hybrid intelligent system: A case study of Iran

This paper concentrates on the effectiveness of using a hybrid intelligent system that combines multilayer perceptron (MLP) neural network, support vector machine (SVM), and logistic regression (LR) classification models with harmony search (HS) optimization algorithm to detect corporate tax evasion for the Iranian National Tax Administration (INTA). In this research, the role of optimization algorithm is to search and find the optimal classification model parameters and financial variables combination. Our proposed system finds optimal structure of the classification model based on the characteristics of the imported dataset. This system has been tested on the data from the food and textile sectors using an iterative structure of 10-fold cross-validation involving 2451 and 2053 test set samples from the tax returns of a two-year period and 1118 and 906 samples as out-of-sample using the tax returns of the consequent year. The results from out-of-sample data show that MLP neural network in combination with HS optimization algorithm outperforms other combinations with 90.07% and 82.45% accuracy, 85.48% and 84.85% sensitivity, and 90.34% and 82.26% specificity, respectively in the food and textile sectors. In addition, there is also a difference between the selected models and obtained accuracies based on the test data and out-of-sample data in both sectors and selected financial variables of every sector.

Normal Tissue Complication Probability (NTCP) Modelling of Severe Acute Mucositis using a Novel Oral Mucosal Surface Organ at Risk

AimsA normal tissue complication probability (NTCP) model of severe acute mucositis would be highly useful to guide clinical decision making and inform radiotherapy planning. We aimed to improve upon our previous model by using a novel oral mucosal surface organ at risk (OAR) in place of an oral cavity OAR. Materials and methodsPredictive models of severe acute mucositis were generated using radiotherapy dose to the oral cavity OAR or mucosal surface OAR and clinical data. Penalised logistic regression and random forest classification (RFC) models were generated for both OARs and compared. Internal validation was carried out with 100-iteration stratified shuffle split cross-validation, using multiple metrics to assess different aspects of model performance. Associations between treatment covariates and severe mucositis were explored using RFC feature importance. ResultsPenalised logistic regression and RFC models using the oral cavity OAR performed at least as well as the models using mucosal surface OAR. Associations between dose metrics and severe mucositis were similar between the mucosal surface and oral cavity models. The volumes of oral cavity or mucosal surface receiving intermediate and high doses were most strongly associated with severe mucositis. ConclusionsThe simpler oral cavity OAR should be preferred over the mucosal surface OAR for NTCP modelling of severe mucositis. We recommend minimising the volume of mucosa receiving intermediate and high doses, where possible.

A New Approach for Mobile Advertising Click-Through Rate Estimation Based on Deep Belief Nets.

In recent years, with the rapid development of mobile Internet and its business applications, mobile advertising Click-Through Rate (CTR) estimation has become a hot research direction in the field of computational advertising, which is used to achieve accurate advertisement delivery for the best benefits in the three-side game between media, advertisers, and audiences. Current research on the estimation of CTR mainly uses the methods and models of machine learning, such as linear model or recommendation algorithms. However, most of these methods are insufficient to extract the data features and cannot reflect the nonlinear relationship between different features. In order to solve these problems, we propose a new model based on Deep Belief Nets to predict the CTR of mobile advertising, which combines together the powerful data representation and feature extraction capability of Deep Belief Nets, with the advantage of simplicity of traditional Logistic Regression models. Based on the training dataset with the information of over 40 million mobile advertisements during a period of 10 days, our experiments show that our new model has better estimation accuracy than the classic Logistic Regression (LR) model by 5.57% and Support Vector Regression (SVR) model by 5.80%.

A comparison of imputation strategies in cluster randomized trials with missing binary outcomes

In cluster randomized trials, clusters of subjects are randomized rather than subjects themselves, and missing outcomes are a concern as in individual randomized trials. We assessed strategies for handling missing data when analysing cluster randomized trials with a binary outcome; strategies included complete case, adjusted complete case, and simple and multiple imputation approaches. We performed a simulation study to assess bias and coverage rate of the population-averaged intervention-effect estimate. Both multiple imputation with a random-effects logistic regression model or classical logistic regression provided unbiased estimates of the intervention effect. Both strategies also showed good coverage properties, even slightly better for multiple imputation with a random-effects logistic regression approach. Finally, this latter approach led to a slightly negatively biased intracluster correlation coefficient estimate but less than that with a classical logistic regression model strategy. We applied these strategies to a real trial randomizing households and comparing ivermectin and malathion to treat head lice.

Discriminating between Native Norway Spruce and Invasive Sitka Spruce—A Comparison of Multitemporal Landsat 8 Imagery, Aerial Images and Airborne Laser Scanner Data

Invasive species can be considered a threat to biodiversity, and remote sensing has been proposed as a tool for detection and monitoring of invasive species. In this study, we test the ability to discriminate between two tree species of the same genera, using data from Landsat 8 satellite imagery, aerial images, and airborne laser scanning. Ground observations from forest stands dominated by either Norway spruce (Picea abies) or Sitka spruce (Picea sitchensis) were coupled with variables derived from each of the three sets of remote sensing data. Random forest, support vector machine, and logistic regression classification models were fit to the data, and the classification accuracy tested by performing a cross-validation. Classification accuracies were compared for different combinations of remote sensing data and classification methods. The overall classification accuracy varied from 0.53 to 0.79, with the highest accuracy obtained using logistic regression with a combination of data derived from Landsat imagery and aerial images. The corresponding kappa value was 0.58. The contribution to the classification accuracy from using airborne data in addition to Landsat imagery was not substantial in this study. The classification accuracy varied between models using data from individual Landsat images.

Geospatial information on geographical and human factors improved anthropogenic fire occurrence modeling in the Chinese boreal forest

We applied a classic logistic regression (LR) model together with a geographically weighted logistic regression (GWLR) model to determine the relationship between anthropogenic fire occurrence and potential driving factors in the Chinese boreal forest and to test whether the explanatory power of the LR model could be increased by considering geospatial information of geographical and human factors using a GWLR model. Three tests, “all variables”, “significant variables”, and “cross-validation”, were applied to compare model performance between the LR and GWLR models. Our results confirmed the importance of distance to railway, elevation, length of fire line, and vegetation cover on fire occurrence in the Chinese boreal forest. In addition, the GWLR model performs better than the LR model in terms of model prediction accuracy, model residual reduction, and spatial parameter estimation by considering geospatial information of explanatory variables. This indicates that the global LR model is incapable of identifying underlying causal factors for wildfire modeling sufficiently. The GWLR model helped identify spatial variation between driving factors and fire occurrence, which can contribute better understanding of forest fire occurrence over large geographic areas and the forest fire management practices may be improved based on it.

Pattern recognition modeling of American English vowel identification by four different identification-proficiency levels of Korean listeners

It has been recently reported through pattern recognition or synthesized vowel perception studies that American English listeners categorize vowels using dynamic spectral properties and duration rather than static spectral properties only (Hillenbrand et al. 1995, Hillenbrand 2013). However, Hong (2015) showed through pattern recognition modeling that Korean listeners used static spectral properties and duration when identifying English vowels. The present study extended Hong (2015) and built a logistic regression classification model to investigate which acoustic cues (static or dynamic spectral features, duration or F0) four different identification-proficiency levels of 133 Korean listeners may use to identify American English monophthongs in /hVd/ syllables. It turned out that the two upper-level groups of Korean listeners used dynamic spectral properties and duration just like American English listeners, whereas the other two lower-level groups used static spectral properties and duration.

系数为LR-型模糊数的模糊Logistic回归模型的参数估计

针对二分类变量问题，经典Logistic回归是合适的。由于观测结果的不精确，响应变量往往介于0，1之间，且没有概率分布，误差也不能完全看作随机性现象。为此，将经典Logistic回归模型与模糊集理论相结合，构建了具有清晰输入–模糊输出的一类模糊Logistic回归模型，其中系数与输出均用LR-型模糊数表示。用成功的可能性替代概率，这些可能性可以由一些语义词描述。然后基于截集构造了模糊数之间的距离，利用此距离得到了上述模型中模糊参数的最小二乘估计。最后将模型应用在狼疮中并通过相容性指数MSI = 0.54说明该模型的有效性。 The classical logistic regression model is appropriate for the problems of binary variable. Since the observations were not crisp value, response variable was often between 0 and 1 and no probability distribution can be considered for response variable; hence error can not be completely regarded as random aspect. Therefore, by combining the classical logistic regression model with fuzzy sets theory, a fuzzy logistic regression model of crisp input and fuzzy output data is constructed; the coefficients and outputs are LR type fuzzy numbers. Considering the possibilities of success instead of the probabilities, the possibilities of success are described by some linguistic terms. Then the distance between two fuzzy numbers is constructed by cut sets. The least squares estimation of fuzzy parameters is obtained in proposed model based on the distance. Finally, the capability index MSI = 0.54 showed that the proposed model is effective of an ordinary one in the modeling Lupus.

Potential for a Simple GPS‐Based Binary Logit Model to Predict Fishing Effort in a Vertical Hook‐and‐Line Reef Fish Fishery

AbstractAccurate fishing effort information is fundamental to the successful management of fisheries resources. Automated, independent, and reliable methods for quantifying fishing effort are needed. The use of vessel speed from Global Positioning System (GPS) data to identify fishing activity has worked well for trawl fisheries but has been less successful in stationary fisheries. Therefore, five trips on four vessels from a vertical hook‐and‐line reef fish fishery were used to examine the efficacy of GPS (speed and time) and electronic video monitoring (EVM) sensor (drum and video) data to corroborate an observer's account of effort using binary logistic regression classification (logit) models as well as a simple speed and time filter (filter). One minute was the minimum data collection interval examined that documented 100% of fishing events. As no fishing occurred at night, opportunistically defined as the 7 h between 2200 and 0500 hours, these records were excluded from analyses. During the day, vessels spent on average 45.2% of the time fishing. Classification success of the approaches examined ranged from 82.4% to 89.5%. Models that included both GPS and EVM sensor data outperformed the filter and GPS‐only models. In general, the filter and most model results can be used as a proxy for observer effort data, at least for the trips examined here. The GPS‐based speed + time logit model was chosen as the preferred approach because of its discriminatory power compared with the filter and the existing widespread use and lower costs of GPS data collection relative to EVM systems and sensors. The speed + time logit model outlined here may have broad utility in this and similar vertical‐line fisheries, including the offshore marine recreational fishing sector.Received September 9, 2015; accepted December 15, 2015

An Application of Bootstrapping in Logistic Regression Model

Computer intensive methods have recently been intensively studied in the field of mathematics, statistics, physics, engineering, behavioral and life sciences. Bootstrap is a computer intensive method that can be used to estimate variability of estimators, estimate probabilities and quantile related to test statistics or to construct confidence intervals, explore the shape of distribution of estimators or test statistics and to construct predictive distributions to show their asymptotic behaviors. In this paper, we fitted the classical logistic regression model, and performed both parametric and non-parametric bootstrap for estimating confidence interval of parameters for logistic model and odds ratio. We also conducted test of hypothesis that the prevalence does not depend on age. Conclusions from both bootstrap methods were similar to those of classical logistic regression.

Cross-species gene expression analysis identifies a novel set of genes implicated in human insulin sensitivity

Objective:Insulin resistance (IR) is one of the earliest predictors of type 2 diabetes. However, diagnosis of IR is limited. High fat fed mouse models provide key insights into IR. We hypothesized that early features of IR are associated with persistent changes in gene expression (GE) and endeavored to (a) develop novel methods for improving signal:noise in analysis of human GE using mouse models; (b) identify a GE motif that accurately diagnoses IR in humans; and (c) identify novel biology associated with IR in humans.Methods:We integrated human muscle GE data with longitudinal mouse GE data and developed an unbiased three-level cross-species analysis platform (single gene, gene set, and networks) to generate a gene expression motif (GEM) indicative of IR. A logistic regression classification model validated GEM in three independent human data sets (n=115).Results:This GEM of 93 genes substantially improved diagnosis of IR compared with routine clinical measures across multiple independent data sets. Individuals misclassified by GEM possessed other metabolic features raising the possibility that they represent a separate metabolic subclass. The GEM was enriched in pathways previously implicated in insulin action and revealed novel associations between β-catenin and Jak1 and IR. Functional analyses using small molecule inhibitors showed an important role for these proteins in insulin action.Conclusions:This study shows that systems approaches for identifying molecular signatures provides a powerful way to stratify individuals into discrete metabolic groups. Moreover, we speculate that the β-catenin pathway may represent a novel biomarker for IR in humans that warrant future investigation.

Estimating bank default with generalised extreme value regression models

The paper proposes a novel model for the prediction of bank failures, on the basis of both macroeconomic and bank-specific microeconomic factors. As bank failures are rare, in the paper we apply a regression method for binary data based on extreme value theory, which turns out to be more effective than classical logistic regression models, as it better leverages the information in the tail of the default distribution. The application of this model to the occurrence of bank defaults in a highly bank dependent economy (Italy) shows that, while microeconomic factors as well as regulatory capital are significant to explain proper failures, macroeconomic factors are relevant only when failures are defined not only in terms of actual defaults but also in terms of mergers and acquisitions. In terms of predictive accuracy, the model based on extreme value theory outperforms classical logistic regression models.

Abstract P4-07-03: MicroRNAs correlating with outcome in patients treated with first-line bevacizumab for metastatic breast cancer

Abstract Background: Biomarkers predicting response to bevacizumab containing therapy in breast cancer are of urgent need. MiRNAs, small non-coding RNAs, can be involved in tumor evolution including regulation of angiogenesis and development of treatment resistance. Therefore, miRNAs can provide both prognostic and predictive information in different cancer entities. Patients and methods: A genome-wide miRNA profiling using high-throughput TaqMan® Array Human MicroRNA Cards enabling quantification of 754 unique human miRNAs was performed. Formalin-fixed paraffin-embedded specimens from 60 patients treated with bevacizumab as first-line therapy at our institution were analyzed. Based on the median overall progression-free survival (PFS) patients were divided into a responder (G1) and a non-responder group (G2). Differentially expressed miRNAs were selected considering a more than two-fold change and a false discovery rate (FDR) < 10% as significant. Further interesting miRNAs were selected by a multivariate logistic regression approach using LASSO (Least Absolute Selection and Shrinkage Operator) regularization. Results: Overall median PFS was 9.3 months with a median PFS of 17.5 and 5.0 months in G1 and G2, respectively. Eight miRNAs (miR-19b-3p, miR-21-5p, miR-9-5p, miR-590-5p, miR-106b-5p, miR-20a-5p) were significantly differentially expressed between these groups (FDR < 10%). Their expression levels were all negatively associated with G1. Additionally, four miRNAs (miR-210-3p, miR-224-5p, miR-155-5p, miR-28-5p) might be interesting predictive biomarkers as they were included in a logistic regression classification model providing an optimal separation between responder and non-responder group (evident from the area under curve [AUC] and 100 x 5-fold cross validation). Currently these data is being validated in a patient cohort not treated with bevacizumab in order to detect the predictive value of these miRNAs for a bevacizumab-based treatment. These validated results will be presented at the meeting. Conclusion: Twelve miRNAs in breast cancer tissue could be identified showing promising predictive value for bevacizumab-based therapy. Although these data need to be confirmed, differences in miRNA expression could help identifying patients with greater benefit from anti-VEGF agents. Citation Format: Simon P Gampenrieder, Frank Hamacher, Gabriel Rinnerthaler, Clemens Hufnagl, Hubert Hackl, Franz Romeder, Claudia Muß, Cornelia Hauser-Kronbeger, Brigitte Mlineritsch, Richard Greil. MicroRNAs correlating with outcome in patients treated with first-line bevacizumab for metastatic breast cancer [abstract]. In: Proceedings of the Thirty-Seventh Annual CTRC-AACR San Antonio Breast Cancer Symposium: 2014 Dec 9-13; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2015;75(9 Suppl):Abstract nr P4-07-03.

Un modelo de regresión logística asimétrico que puede explicar la probabilidad de éxito en el rendimiento académico

En este estudio se pretende explicar la probabilidad de éxito en la asignatura Matemáticas Empresariales correspondiente al Grado en Administración y Dirección de Empresas en la Universidad de Las Palmas de Gran Canaria. Para ello se intentará detectar qué factores influyen en dicha probabilidad utilizando, primero un modelo de regresión logística clásico y, en segundo lugar, un modelo de regresión logística asimétrico Bayesiano. Los resultados obtenidos permiten concluir que las variables significativas que podrían determinar el rendimiento académico en términos de la probabilidad relativa de aprobar la asignatura son: la asistencia con regularidad a clases de teoría y prácticas, que el estudiante valore positivamente el material del que dispone para el seguimiento de la asignatura, el tipo de centro en que se cursaron los estudios preuniversitarios y la asistencia a clases de apoyo. La discusión abre líneas de trabajo futuras que analicen las relaciones entre algunas de las variables consideradas.<!--[endif] -->

A novel logistic multi-class supervised classification model based on multi-fractal spectrum parameters for hyperspectral data

A novel logistic multi-class supervised classification model based on multi-fractal spectrum parameters is proposed to avoid the error that is caused by the difference between the real data distribution and the hypothetic Gaussian distribution and avoid the computational burden working in the logistic regression classification directly for hyperspectral data. The multi-fractal spectra and parameters are calculated firstly with training samples along the spectral dimension of hyperspectral data. Secondly, the logistic regression model is employed in our work because the logistic regression classification model is a distribution-free nonlinear model which is based on the conditional probability without the Gaussian distribution assumption of the random variables, and the obtained multi-fractal parameters are applied to establish the multi-class logistic regression classification model. Finally, the Newton–Raphson method is applied to estimate the model parameters via the maximum likelihood algorithm. The classification results of the proposed model are compared with the logistic regression classification model based on an adaptive bands selection method by using the Airborne Visible/Infrared Imaging Spectrometer and airborne Push Hyperspectral Imager data. The results illuminate that the proposed approach achieves better accuracy with lower computational cost simultaneously.

Natal origin of black ducks harvested during the sport hunt in Québec, Canada

In Québec, Canada, the American black duck (Anas rubripes) breeds mainly in the boreal forest to the north, but a small segment of the population breeds in southern Québec in the densely populated agricultural areas surrounding the St. Lawrence River and Estuary. This agricultural breeding population faces increasing pressure from urbanization, competition with mallards (A. platyrhynchos), landuse changes, and hunting. In Québec, approximately 69% of black ducks are harvested in agricultural landscapes. Most birds harvested in Québec are taken in September, during the first 3 weeks of the hunting season, when birds breeding in the boreal habitat are unlikely to have initiated their southbound fall migration. This additional pressure on the agricultural breeding population could be detrimental to its long-term sustainability. We measured carbon (δ13C), nitrogen (δ15N), hydrogen (δD), and sulfur (δ34S) isotopic compositions of feathers from hatching-year black ducks of known natal origin (banded birds) and used these to develop a multivariate logistic regression classification model based on habitat type. We then applied this model to feathers from birds harvested in Québec during the fall sport hunt in those same years to determine their natal origin. Sixty percent in 2004 and 17% in 2005 of black ducks harvested during the fall sport hunt in Québec originated from the agricultural segment of the population, compared to 36% and 72% in those same years originating from the boreal area. This is in stark contrast to the species' distribution in Québec, with only approximately 4% of the population breeding in the agricultural portion of the province. Our results suggest that this harvest bias toward the agricultural segment of the breeding population, which may be accentuated in years when migration is delayed, could compromise its long-term sustainability. We highlight the importance of existing regional survey programs, and the information gaps that exist concerning the natal origin of black ducks harvested in the northeastern United States, and the annual productivity of the agricultural black duck population in Québec. Stable isotopes from samples of known natal origin (banded birds) proved an effective tool to identify the natal origin of hunted birds. However, we caution against the use of published isotopic thresholds (δ13C, δ15N) to infer the origin of migrating organisms in small study areas. © 2014 The Wildlife Society.

Synergistic association between two alcohol metabolism relevant genes and coronary artery disease among Chinese hypertensive patients.

ObjectiveCoronary artery disease (CAD) is a multifactorial and polygenic disease. The aim of this study was to examine the association between six polymorphisms of four alcohol metabolism relevant genes (ADH1B, ADH1C, ALDH1b1, ALDH2) and the risk of CAD in Han Chinese.Methods and ResultsThis was a hospital-based case-control study involving 1365 hypertensive patients. All study subjects were angiographically confirmed. Genotypes were determined with ligase detection reaction method. There was no observable deviation from the Hardy-Weinberg equilibrium for six examined polymorphisms in controls. The genotype and allele distributions of ALDH1b1 rs2073478 and ALDH2 rs671 polymorphisms differed significantly between the two groups (P≤0.005), even after the Bonferroni correction. The most common allele combination was A-C-C-G-C-G (alleles in order of rs1229984, rs1693482, rs2228093, rs2073478, rs886205, rs671) and its frequency was slightly higher in controls than in CAD patients (P = 0.067). After assigning the most common allele combination as a reference, allele combination A-C-C-T-C-A, which simultaneously possessed the risk alleles of rs2073478 and rs671 polymorphisms, was associated with a 1.80-fold greater risk of CAD. Further, a two-locus model including rs2073478 and rs671 that had a maximal testing accuracy of 0.598 and a cross-validation consistency of 10 (P = 0.008) was deemed as the overall best MDR model, which was further validated by classical Logistic regression model.ConclusionOur findings provide clear evidence for both individual and interactive associations of ALDH1b1 and ALDH2 genes with the development of CAD in Han Chinese.

Signatures for Mass Spectrometry Data Quality

Ensuring data quality and proper instrument functionality is a prerequisite for scientific investigation. Manual quality assurance is time-consuming and subjective. Metrics for describing liquid chromatography mass spectrometry (LC–MS) data have been developed; however, the wide variety of LC–MS instruments and configurations precludes applying a simple cutoff. Using 1150 manually classified quality control (QC) data sets, we trained logistic regression classification models to predict whether a data set is in or out of control. Model parameters were optimized by minimizing a loss function that accounts for the trade-off between false positive and false negative errors. The classifier models detected bad data sets with high sensitivity while maintaining high specificity. Moreover, the composite classifier was dramatically more specific than single metrics. Finally, we evaluated the performance of the classifier on a separate validation set where it performed comparably to the results for the testing/training data sets. By presenting the methods and software used to create the classifier, other groups can create a classifier for their specific QC regimen, which is highly variable lab-to-lab. In total, this manuscript presents 3400 LC–MS data sets for the same QC sample (whole cell lysate of Shewanella oneidensis), deposited to the ProteomeXchange with identifiers PXD000320–PXD000324.

A novel model enhances HbA1c‐based diabetes screening using simple anthropometric, anamnestic, and demographic information 一个新颖的使用简单人体测量学参数、既往史以及人口统计学信息增强基于HbA1c检查效力的糖尿病筛查模型

AbstractBackgroundThe sensitivity of HbA1c is not optimal for the screening of patients with latent diabetes. We hypothesize that simple healthcare information could improve accuracy.MethodsWe retrospectively analyzed data, including HbA1c, from multiple years from the National Health and Nutrition Examination Survey (NHANES) database (2005–2010). The data were used to create a logistic regression classification model for screening purposes.ResultsThe study evaluated data for 5381 participants, including 404 with undiagnosed diabetes. The HbA1c screening data were supplemented with information about age, waist circumference, and physical activity in the HbA1c+ model. Alone, HbA1c alone had a receiver operating characteristics (ROC) curve for the area under the curve (AUC) of 0.808 (95% confidence interval [CI] 0.792–0.834). The HbA1c+ model had an ROC AUC of 0.851 (95% CI 0.843–0.872). There was a significant difference in the AUC between our model and using HbA1c without supplementary information (P < 0.05).ConclusionsWe have developed a novel screening model that could help improve screening for type 2 diabetes with HbA1c. It seems beneficial to systematically add additional patient healthcare information in the process of screening with HbA1c.