Boosted Tree Classifier Research Articles

Fingerprint liveness detection using local quality features

Fingerprint-based recognition is widely deployed in different domains. However, current recognition systems are vulnerable to presentation attack. Presentation attack utilizes an artificial replica of a fingerprint to deceive the sensors. In such scenarios, fingerprint liveness detection is required to ensure the actual presence of a live fingerprint. In this paper, we propose a static software-based approach using quality features to detect the liveness in a fingerprint image. The proposed method extracts eight sensor-independent quality features from the detailed ridge–valley structure of a fingerprint at the local level to form a 13-dimensional feature vector. Sequential Forward Floating Selection and Random Forest Feature Selection are used to select the optimal feature set from the created feature vector. To classify fake and live fingerprints, we have used support vector machine, random forest, and gradient boosted tree classifiers. The proposed method is tested on a publically available database of LivDet 2009 competition. The experimental results demonstrate that the least average classification error of 5.3% is achieved on LivDet 2009 database, exhibiting supremacy of the proposed method over current state-of-the-art approaches. Additionally, we have analyzed the importance of individual features on LivDet 2009 database, and effectiveness of the best-performing features is evaluated on LivDet 2011, 2013, and 2015 databases. The obtained results depict that the proposed approach is able to perform well irrespective of the different sensors and materials used in these databases. Further, the proposed method utilizes a single fingerprint image. This characteristic makes our method more user-friendly, faster, and less intrusive.

Machine-learned models using hematological inflammation markers in the prediction of short-term acute coronary syndrome outcomes

BackgroundIncreased systemic and local inflammation play a vital role in the pathophysiology of acute coronary syndrome. This study aimed to assess the usefulness of selected machine learning methods and hematological markers of inflammation in predicting short-term outcomes of acute coronary syndrome (ACS).MethodsWe analyzed the predictive importance of laboratory and clinical features in 6769 hospitalizations of patients with ACS. Two binary classifications were considered: significant coronary lesion (SCL) or lack of SCL, and in-hospital death or survival. SCL was observed in 73% of patients. In-hospital mortality was observed in 1.4% of patients and it was higher in the case of patients with SCL. Ensembles of decision trees and decision rule models were trained to predict these classifications.ResultsThe best performing model for in-hospital mortality was based on the dominance-based rough set approach and the full set of laboratory as well as clinical features. This model achieved 81 ± 2.4% sensitivity and 81.1 ± 0.5% specificity in the detection of in-hospital mortality. The models trained for SCL performed considerably worse. The best performing model for detecting SCL achieved 56.9 ± 0.2% sensitivity and 66.9 ± 0.2% specificity. Dominance rough set approach classifier operating on the full set of clinical and laboratory features identifies presence or absence of diabetes, systolic and diastolic blood pressure and prothrombin time as having the highest confirmation measures (best predictive value) in the detection of in-hospital mortality. When we used the limited set of variables, neutrophil count, age, systolic and diastolic pressure and heart rate (taken at admission) achieved the high feature importance scores (provided by the gradient boosted trees classifier) as well as the positive confirmation measures (provided by the dominance-based rough set approach classifier).ConclusionsMachine learned models can rely on the association between the elevated inflammatory markers and the short-term ACS outcomes to provide accurate predictions. Moreover, such models can help assess the usefulness of laboratory and clinical features in predicting the in-hospital mortality of ACS patients.

Macro‐spatial structure of biotic interactions in the distribution of a raptor species

AbstractAimWhile the contribution of abiotic factors to species distribution is well known, the geographic structure, if any, of biotic interactions within the species range is poorly understood. Most studies neglect biotic interactions when generating Species Distribution Models (SDMs) and projecting them, using future climatic scenarios, while others argue that biotic interactions may extend species tolerances to suboptimal abiotic conditions. Elucidating the extent to which biotic interactions play a role at a macro‐scale is challenging due to its inherent complexity. In this study, we characterized the independent contribution of prey abundance distributions on the Merlin's wintering distribution (Falco columbarius). Then, to examine the hypothesis that biotic interactions may counteract other suboptimal conditions, we tested for a differential importance of physical habitat characteristics and prey abundance distributions along the species’ wintering range.LocationPeninsular Spain.MethodsWe modelled the Merlin's geographic distribution with Boosted Classification Trees as a function of environmental predictors (environmental model) and prey relative abundances (prey model), either separately or jointly (combined model). We tested whether the predictive success of environmental and prey models differ spatially.ResultsPartialling out the variation into independent components we found that the prey abundance distributions explained the largest part of variation. Furthermore, the first four predictors with the highest contribution in our combined models were the abundances of prey species. Finally, our model predictions revealed a north‐to‐south increase in the importance of prey abundance distributions. Interestingly, our results suggest that biotic interactions can enable species to inhabit a wider range of suboptimal habitat conditions on range margins.Main conclusionsRelevant biotic interactions may not be always fully interchangeable with environmental surrogates. Abiotic factors and biotic interactions may shape species range limits and cores of distributions differently. Neglecting biotic interactions may compromise spatiotemporal transferability of SDMs, especially on species margins, and hence their applicability.

Application of Feature Extraction and Classification Methods for Histopathological Image using GLCM, LBP, LBGLCM, GLRLM and SFTA

Classification of histopathologic images and identification of cancerous areas is quite challenging due to image background complexity and resolution. The difference between normal tissue and cancerous tissue is very small in some cases. So, the features of the tissue patches in the image have key importance for automatic classification. Using only one feature or using a few features leads to poor classification results because of the small difference between the textures. In this study, the classification results are compared using different feature extraction algorithms that can extract various features from histopathological image texture. For this study, GLCM, LBP, LBGLCM, GLRLM and SFTA algorithms which are successful feature extraction algorithms have been chosen. The features obtained from these methods are classified with SVM, KNN, LDA and Boosted Tree classifiers. The most successful feature extraction algorithm for histopathological images is determined and the most successful classification algorithm is determined.

Fast Local Laplacian‐Based Steerable and Sobel Filters Integrated with Adaptive Boosting Classification Tree for Automatic Recognition of Asphalt Pavement Cracks

Effective road maintenance requires adequate periodic surveys of asphalt pavement condition. The manual process of pavement assessment is labor intensive and time‐consuming. This study proposes an alternative for automating the periodic surveys of pavement condition by means of image processing and machine learning. Advanced image processing techniques including fast local Laplacian filter, Sobel filter, steerable filter, and projection integral are employed for image enhancement and analysis to extract useful features from digital images. Based on the features produced by these image processing techniques, adaptive boosting classification tree is used to perform pavement crack recognition tasks. A dataset of image samples consisting of five classes (alligator crack, diagonal crack, longitudinal crack, noncrack, and transverse crack) has been collected to construct and verify the performance of the adaptive boosting classification tree. The experimental results show that the proposed approach has achieved a high crack classification accuracy which is roughly 90%. Therefore, the newly developed model is a promising alternative to help transportation agencies in pavement condition evaluation.

Boosted Classification Trees and Class Probability/Quantile Estimation

The standard by which binary classifiers are usually judged, misclassification error, assumes equal costs of misclassifying the two classes or, equivalently, classifying at the 1/2 quantile of the conditional class probability function P[y=1|x]. Boosted classification trees are known to perform quite well for such problems. In this article we consider the use of standard, off-the-shelf boosting for two more general problems: 1) classification with unequal costs or, equivalently, classification at quantiles other than 1/2, and 2) estimation of the conditional class probability function P[y=1|x]. We first examine whether the latter problem, estimation of P[y=1|x], can be solved with LogitBoost, and with AdaBoost when combined with a natural link function. The answer is negative: both approaches are often ineffective because they overfit P[y=1|x] even though they perform well as classifiers. A major negative point of the present article is the disconnect between class probability estimation and classification. Next we consider the practice of over/under-sampling of the two classes. We present an algorithm that uses AdaBoost in conjunction with Over/Under-Sampling and Jittering of the data JOUS-Boost. This algorithm is simple, yet successful, and it preserves the advantage of relative protection against overfitting, but for arbitrary misclassification costs and, equivalently, arbitrary quantile boundaries. We then use collections of classifiers obtained from a grid of quantiles to form estimators of class probabilities. The estimates of the class probabilities compare favorably to those obtained by a variety of methods across both simulated and real data sets.

Classifying symmetrical differences and temporal change for the detection of malignant masses in mammography using deep neural networks.

We investigate the addition of symmetry and temporal context information to a deep convolutional neural network (CNN) with the purpose of detecting malignant soft tissue lesions in mammography. We employ a simple linear mapping that takes the location of a mass candidate and maps it to either the contralateral or prior mammogram, and regions of interest (ROIs) are extracted around each location. Two different architectures are subsequently explored: (1)a fusion model employing two datastreams where both ROIs are fed to the network during training and testing and (2)a stagewise approach where a single ROI CNN is trained on the primary image and subsequently used as a feature extractor for both primary and contralateral or prior ROIs. A "shallow" gradient boosted tree classifier is then trained on the concatenation of these features and used to classify the joint representation. The baseline yielded an AUC of 0.87 with confidence interval [0.853, 0.893]. For the analysis of symmetrical differences, the first architecture where both primary and contralateral patches are presented during training obtained an AUC of 0.895 with confidence interval [0.877, 0.913], and the second architecture where a new classifier is retrained on the concatenation an AUC of 0.88 with confidence interval [0.859, 0.9]. We found a significant difference between the first architecture and the baseline at high specificity with [Formula: see text]. When using the same architectures to analyze temporal change, we yielded an AUC of 0.884 with confidence interval [0.865, 0.902] for the first architecture and an AUC of 0.879 with confidence interval [0.858, 0.898] in the second setting. Although improvements for temporal analysis were consistent, they were not found to be significant. The results show our proposed method is promising and we suspect performance can greatly be improved when more temporal data become available.

Habitat suitability-density relationship in an endangered woodland species: the case of the Blue Chaffinch (Fringilla polatzeki).

BackgroundUnderstanding constraints to the distribution of threatened species may help to ascertain whether there are other suitable sectors for reducing the risks associated with species that are recorded in only one protected locality, and to inform about the suitability of other areas for reintroduction or translocation programs.MethodsWe studied the Gran Canaria blue chaffinch (Fringilla polatzeki), a habitat specialist endemic of the Canary Islands restricted to the pine forest of Inagua, the only area where the species has been naturally present as a regular breeder in the last 25 years. A suitability distribution model using occurrences with demographic relevance (i.e., nest locations of successful breeding attempts analysed using boosted classification trees) was built considering orographic, climatic and habitat structure predictors. By means of a standardized survey program we monitored the yearly abundance of the species in 100 sectors since the declaration of Inagua as a Strict Nature Reserve in 1994.ResultsThe variables with the highest relative importance in blue chaffinch habitat preferences were pine height, tree cover, altitude, and rainfall during the driest trimester (July–September). The observed local abundance of the blue chaffinch in Inagua (survey data) was significantly correlated with habitat suitability derived from modelling the location of successful nesting attempts (using linear and quantile regressions). The outcomes of the habitat suitability model were used to quantify the suitability of other natural, historic, pine forests of Gran Canaria. Tamadaba is the forest with most suitable woodland patches for the species. We estimated a population size of 195–430 blue chaffinches in Inagua since 2011 (95% CI), the smallest population size of a woodland passerine in the Western Palearctic.DiscussionHabitat suitability obtained from modelling the location of successful breeding attempts is a good surrogate of the observed local abundance during the reproductive season. The outcomes of these models can be used for the identification of potential areas for the reintroduction of the species in other suitable pine forests and to inform forest management practices.

Detection of ventricular tachycardia and fibrillation using adaptive variational mode decomposition and boosted-CART classifier

Abstract Rapid ventricular tachycardia (VT) and ventricular fibrillation (VF) are serious life-threatening ventricular arrhythmias. Correct detection of VT/VF is crucial for the rescue of cardiac arrest patient. In this paper, we proposed a new method for improving the detection effect of VT/VF. An adaptive variational mode decomposition (adaptive-VMD) algorithm was presented to decompose the electrocardiogram (ECG) signal into five band-limited intrinsic modes (BLIMs). Then, a total of 6 features were extracted from these BLIMs to characterize the details of VT/VF. Last, a boosted classification and regression tree (Boosted-CART) classifier that combines feature selection and recognition was used to detect VT/VF. Three annotated public ECG databases were used as the training and testing datasets. Ten-fold cross-validation was implemented to assess the performance of the method. An accuracy (Acc) of 98.29% ± 0.18%, a sensitivity (SE) of 97.32% ± 0.12% and a specificity (SP) of 98.95% ± 0.84% were obtained. In comparison with the existing state-of-the-art methods for VT/VF detection, the proposed method demonstrated better overall performance.

Neural Network Classifier for Automatic Detection of Invasive Versus Noninvasive Airway Management Technique Based on Respiratory Monitoring Parameters in a Pediatric Anesthesia.

Children undergoing general anesthesia require airway monitoring by an anesthesia provider. The airway may be supported with noninvasive devices such as face mask or invasive devices such as a laryngeal mask airway or an endotracheal tube. The physiologic data stored provides an opportunity to apply machine learning algorithms distinguish between these modes based on pattern recognition. We retrieved three data sets from patients receiving general anesthesia in 2015 with either mask, laryngeal mask airway or endotracheal tube. Patients underwent myringotomy, tonsillectomy, adenoidectomy or inguinal hernia repair procedures. We retrieved measurements for end-tidal carbon dioxide, tidal volume, and peak inspiratory pressure and calculated statistical features for each data element per patient. We applied machine learning algorithms (decision tree, support vector machine, and neural network) to classify patients into noninvasive or invasive airway device support. We identified 300 patients per group (mask, laryngeal mask airway, and endotracheal tube) for a total of 900 patients. The neural network classifier performed better than the boosted trees and support vector machine classifiers based on the test data sets. The sensitivity, specificity, and accuracy for neural network classification are 97.5%, 96.3%, and 95.8%. In contrast, the sensitivity, specificity, and accuracy of support vector machine are 89.1%, 92.3%, and 88.3% and with the boosted tree classifier they are 93.8%, 92.1%, and 91.4%. We describe a method to automatically distinguish between noninvasive and invasive airway device support in a pediatric surgical setting based on respiratory monitoring parameters. The results show that the neural network classifier algorithm can accurately classify noninvasive and invasive airway device support.

Machine learning to identify multigland disease in primary hyperparathyroidism

Background20%-25% of patients with primary hyperparathyroidism will have multigland disease (MGD). Preoperatative imaging can be inaccurate or unnecessary in MGD. Identification of MGD could direct the need for imaging and inform operative approach. The purpose of this study is to use machine learning (ML) methods to predict MGD. MethodsRetrospective review of a prospective database. The ML platform, Waikato Environment for Knowledge Analysis, was used, and we selected models for (1) overall accuracy and (2) preferential identification of MGD. A review of imaging studies was performed on a cohort predicted to have MGD. Results2010 patients met inclusion criteria: 1532 patients had single adenoma (SA) (76%) and 478 had MGD (24%). After testing many algorithms, we selected two different models for potential integration as clinical decision-support tools. The best overall accuracy was achieved using a boosted tree classifier, RandomTree: 94.1% accuracy; 94.1% sensitivity, 83.8% specificity, 94.1% positive predictive value, and 0.984 area under the receiver operating characteristics curve. To maximize positive predictive value of MGD prediction, a rule-based classifier, JRip, with cost-sensitive learning was used and achieved 100% positive predictive value for MGD. Imaging reviewed from the cohort of 34 patients predicted to have MGD by the cost-sensitive model revealed 39 total studies performed: 28 sestamibi scans and 11 ultrasounds. Only 8 (29%) sestamibi scans and 4 (36%) ultrasounds were correct. ConclusionsML methods can help distinguish MGD early in the clinical evaluation of primary hyperparathyroidism, guiding further workup and surgical planning.

Applying machine learning on sensor data for irrigation recommendations: revealing the agronomist’s tacit knowledge

Jojoba Israel is a world-leading producer of Jojoba products, whose orchards are covered with sensors that collect soil moisture data for monitoring plant needs at real-time. Based on these data, the company’s agronomist defines a weekly irrigation plan. In addition, data on weather, irrigation, and yield are recorded from other sources (e.g. meteorological station and irrigation-plan records). However, so far, there has been no attempt to use the entire set of collected data to reveal insights and interesting relationships between different variables, such as soil, weather, irrigation characteristics, and resulting yield. By integrating and utilizing data from different sources, our research aims at using the collected data not only for monitoring and controlling the crop, but also for predicting irrigation recommendations. In particular, a dataset was constructed by integrating data collected over almost two years from 22 soil-sensors spread in four major plots (which are divided into 28 subplots and eight irrigation groups), from a meteorological station, and from actual irrigation records. Different regression and classification algorithms were applied on this dataset to develop models that were able to predict the weekly irrigation plan as recommended by the agronomist. The models were developed using eight different subsets of variables to determine which variables consistently contributed to prediction accuracy. By comparing the resulting models, it was shown that the best regression model was Gradient Boosted Regression Trees, with 93% accuracy, and the best classification model was the Boosted Tree Classifier, with 95% accuracy (on the test-set). Data that were not contributing to the model prediction success rate were identified as well. The resulting model can significantly facilitate the agronomist’s irrigation planning process. In addition, the potential of applying machine learning on the company data for yield and disease prediction is discussed.

A data-driven approach to assess large fire size generation in Greece

Identifying factors and drivers which control large fire size generation is critical for planning fire management activities. This study attempts to determine the role of fire suppression tactics and behavior, weather, topography and landscape features on two different datasets of large fire size (500–1000 ha) and very large fire size (>1000 ha) compared to two datasets of small fire size (<50 ha) which occurred in Greece, during the period 1984–2009. In this context, we used a logistic regression (LR) analysis and two machine learning algorithms: random forest (RF) and boosting classification trees (BCT). The models comparison was based on the area under the receiver operating characteristic curve and the observed overall accuracy. The comparison indicated that RF had greater ability than LR and BCT to predict large fire generation. Results from the RF classifier algorithm showed that large fire generation mainly depended on fire suppression tactics and the prevailing weather conditions. This improved understanding of the factors which drive fire ignitions to turn into large fire sizes will provide the opportunity for land and forest managers to increase fire awareness and the development of concrete initiatives for successful fire management.

The refined ORCS approach: A catch-based method for estimating stock status and catch limits for data-poor fish stocks

The ‘Only Reliable Catch Stocks’ (ORCS) Working Group approach to data-poor fisheries stock status and catch limit estimation has been used by U.S. fisheries managers but has yet to be fully evaluated. The ORCS approach estimates stock status using a fourteen question ‘Table of Attributes’ and the overfishing limit by multiplying a historical catch statistic by a scalar based on the estimated status. We evaluated the performance of the approach by applying it to 193 stocks with data-rich stock assessments and comparing its predictions of stock status with the assessment model estimates. The approach classified all but three stocks as fully exploited indicating that it is a poor predictor of status and should not be used by managers. We refined the original ORCS approach by: (1) developing a more predictive model of stock status using boosted classification trees and (2) identifying the historical catch statistics and scalars that best estimate overfishing limits using assessment model data. The refined ORCS approach correctly classified 74% of all stocks and 62% of overexploited stocks in a training dataset and 74% of all stocks and 50% of overexploited stocks in an independent test dataset. The refined approach performed better than six other widely used catch-only methods. However, the overfishing limits estimated by the refined approach would further deplete overexploited stocks without the use of conservative catch scalars to buffer against classification uncertainty. Conservative catch scalars can reduce the probability of overfishing below 50%, the U.S. legal maximum, but with concomitant increases in the probability and magnitude of underfishing. The refined ORCS approach may therefore be useful when other methods are not possible or appropriate and some risk of underfishing is acceptable.

Monitoring fatigue in construction workers using physiological measurements

Fatigue is one of the factors leading to reduction in productivity, poor quality of work and increased risk of accidents in construction. Existing established methods of assessing fatigue include surveys and questionnaires, which are cumbersome to implement at construction sites. This study presents a novel approach for real time monitoring of physical fatigue in construction workers using wearable sensors. Changes in the heart rate, thermoregulation and electrical brain activity during a simulated construction task were monitored from 12 participants using a heart rate monitor, infrared temperature sensors and an EEG sensor. Borg's RPE was used as a subjective scale to collect the level of fatigue experienced by the participants. Boosted tree classifiers were trained using the features extracted from the heart rate and temperature sensor signals and used to predict the level of physical fatigue. Only physical fatigue was assessed as none of the participants developed any sign of mental fatigue during the study. The results show that physical fatigue can be monitored using wearable sensors. The classification accuracy, based solely on features extracted from average of skin temperature data, was 9% higher than based solely on heart rate data, and combining the information from both sensors resulted in the best accuracy of 82%. The results also show that monitoring thermoregulation from temple can be more useful compared to other studied monitoring sites, the classification accuracy based only on data from the temple was 79%. This accuracy is significantly higher compared to the classification accuracy based only on heart rate data (59%).

Recognition of culling reasons in Polish dairy cows using data mining methods

Cow longevity and reasons for culling are one of the most important research problems in the contemporary cattle breeding. Therefore, the analysis of the relationship between cow performance and involuntary disposal contributes to taking more informed decisions in herd management. The aim of the present study was to compare the efficiency of artificial neural networks (ANN) and boosted classification trees (BT) with that of linear discriminant analysis (LDA) and classification functions (CF) in recognizing culling reasons of dairy cows in Poland, based on the lifetime performance data, routinely monitored in a herd. The analyses carried out in the present study showed that the accurate recognition of different culling reasons based on predictors included in the above-mentioned models is, in general, impossible. Only BT had limited discrimination abilities, but the results obtained using this method were not much improved compared with ANN and LDA with CF. In order to predict precisely various culling reasons, more specific data are required. They could be obtained from the increasingly popular, technologically advanced, systems of real-time monitoring of animal health status (physical activity, rumination rate, etc.), dependent also on environmental conditions (e.g. temperature–humidity index).

Selecting site characteristics at different spatial and thematic scales for shrubby cinquefoil (Potentilla fruticosa L.) distribution mapping

Abstract The largest natural population of shrubby cinquefoil (Potentilla fruticosa) in the Baltic States was observed in the field to reveal the scale-dependent explanatory value of site characteristics for subsequent spatial distribution modelling of the species. About 700 km was crossed during field observations in 2008–2014. Thinning of the raw field records to ensure a distance of at least 50 metres between each point yielded 1459 presences and 7327 absences. These occurrence data were related to present and historical land cover, soil, elevation, human population density, the proportion of presence sites, and P. fruticosa mean coverage in the neighbourhood. Boosted classification tree models were used to compare the value of 60 individual site features at thematically and spatially different levels of generalization as indicators of the species’ presence or absence. P. fruticosa presence is significantly non-random regarding most of the studied site features but only a few of these are valuable predictors. The proportion of presences in the neighbourhood had the highest indicative value. P. fruticosa occurrence also coincides with moist thin calcareous soils according to the soil map, with larger scrubland patches according to the topographical database, and with tussock areas according to a topographical map from the 1930s. The explanatory value of nominal site characteristics primarily drops when the most indicative category is merged with other classes to form a more general category. Site characteristics calculated at the observation point are not always the most effective predictors for P. fruticosa occurrence – features of the neighbourhood are related to the occurrence as well. The study area was classified into: confirmed absence area, unclear presence/absence area and probable presence area. Subsequent distribution modelling in the unclear area should be targeted on a species presence/absence, while abundance could be the priority within the probable presence area.

Banking failure prediction: a boosting classification tree approach

ABSTRACTThe recent financial crisis shows that failure of some financial institutions can cause other banks to fail and ultimately cause damage to the financial system worldwide. Eurozone banks that experienced either liquidity or solvency problems during the financial markets turmoil were bailed out by their national governments with the financial support and supervision of the European Union. This paper applies the boosted classification tree methodology to predict failure in the banking sector and identifies four key scorecard variables that are worth tracking closely in order to anticipate and prevent bank financial distress. The data used in this study comprises 2006–2012 annual series of 25 financial ratios of 155 banks in the Eurozone. The findings indicate that the greater the size and the higher the income from non-operating items and net loans to deposits, the more likely is bank failure; conversely, the higher the Interbank ratio the lower the chances of bank financial distress. For the sake of their own financial soundness, banks should fund lending activities through clients’ deposits and should avoid relying excessively on non-recurring sources of income.

Predicting regional densities from bird occurrence data: validation and effects of species traits in a Macaronesian Island

AbstractAimQuantifying species abundances is costly, especially when many species are involved. To overcome this problem, several studies have predicted local abundances (at the sample unit level) from species occurrence distribution models (SODMs), with differences in predictive performance among studies. Surprisingly, the ability of SODM to predict regional abundances of an entire area of interest has never been tested, despite the fact that it is an essential parameter for species conservation and management. We tested whether local and regional abundances of 21 terrestrial bird species could be predicted from SODMs in an exhaustively surveyed island, and examined the variation explained by species‐specific traits.LocationLa Palma Island, Canary Islands.MethodsWe firstly assessed two types of algorithms representing the two main families of SODMs. We built models using presence/absence (boosted classification trees) and presence/background (MaxEnt) data as a function of relevant environmental predictors and tested their ability to predict the observed local abundances. The predicted probabilities of occurrence (Pi) were translated into animal numbers (n′) using the revisited equation ni′ = −ln(1−Pi), and we obtained regional abundances (for the whole island).ResultsPredictive ability of presence/absence models was superior than that of MaxEnt. At the regional level, the observed average densities of all species were highly predictable from occurrence probabilities (R2 = 93.5%), without overall overestimation or underestimation. Interspecific variation in the accuracy of predicted regional density was largely explained (R2 = 73%), with habitat breath and variation in local abundance being the traits of greatest importance.Main conclusionsDespite uncertainties associated with local predictions and the idiosyncrasies of each species, our procedures enabled us to predict regional abundances in an unbiased way. Our approach provides a cost‐effective tool when a large number of species are involved. Furthermore, the influence of species‐specific traits on the prediction accuracy provides insights into sampling designs for focal species.

Statistical analysis with missing exposure data measured by proxy respondents: a misclassification problem within a missing-data problem.

In studies of older adults, researchers often recruit proxy respondents, such as relatives or caregivers, when study participants cannot provide self-reports (e.g., because of illness). Proxies are usually only sought to report on behalf of participants with missing self-reports; thus, either a participant self-report or proxy report, but not both, is available for each participant. Furthermore, the missing-data mechanism for participant self-reports is not identifiable and may be nonignorable. When exposures are binary and participant self-reports are conceptualized as the gold standard, substituting error-prone proxy reports for missing participant self-reports may produce biased estimates of outcome means. Researchers can handle this data structure by treating the problem as one of misclassification within the stratum of participants with missing self-reports. Most methods for addressing exposure misclassification require validation data, replicate data, or an assumption of nondifferential misclassification; other methods may result in an exposure misclassification model that is incompatible with the analysis model. We propose a model that makes none of the aforementioned requirements and still preserves model compatibility. Two user-specified tuning parameters encode the exposure misclassification model. Two proposed approaches estimate outcome means standardized for (potentially) high-dimensional covariates using multiple imputation followed by propensity score methods. The first method is parametric and uses maximum likelihood to estimate the exposure misclassification model (i.e., the imputation model) and the propensity score model (i.e., the analysis model); the second method is nonparametric and uses boosted classification and regression trees to estimate both models. We apply both methods to a study of elderly hip fracture patients.