Terms Of Misclassification Cost Research Articles

Development of a Nomogram to Estimate the 60-Day Probability of Death or Culling Due to Severe Clinical Mastitis in Dairy Cows at First Veterinary Clinical Evaluation

Severe clinical mastitis is a frequent disease of dairy cattle. An effective mean of predicting survival despite treatment would be helpful for making euthanasia decisions in poor prognosis cases. The objective was to develop a nomogram for prediction of death or culling in the 60 days following a severe mastitis episode in dairy cows at first veterinary visit in farm settings. A total of 224 dairy cows presenting severe clinical mastitis and examined for the first time by a veterinarian were included in a prospective study. Clinical and laboratory (complete blood cell count, L-lactate, cardiac troponin I, milk culture) variables were recorded. Animals were followed for 60 days. A nomogram was built with an adaptive elastic-net Cox proportional hazards model. Performances and relevance were evaluated by area under the receiver operating characteristic curve (AUC), Harrell's concordance index (C-index), calibration curve, decision curve analysis (DCA) and misclassification cost term (MCT). The nomogram included: lactation number, recumbency, depression intensity, capillary refilling time, ruminal motility rate, dehydration level, lactates concentration, hematocrit, band neutrophils count, monocyte count, and milk bacteriology. The AUC and C-index showed a good calibration and ability to discriminate. The DCA suggested that the nomogram was clinically relevant. Euthanizing animals having less than 25% probability of survival is economically optimal. It could be used for early euthanasia decisions in animals that would not survive despite treatment. To facilitate the use of this nomogram by veterinarians, a web-based app was developed.

Bayesian accuracy estimates and fit for purpose thresholds of cytology and culture of endometrial swab samples for detecting endometritis in mares

The overall aim of this work was to identify the potential impact of misclassification errors associated with routine screening and diagnostic testing for endometritis in mares. Using Bayesian latent class models (BLCM), specific objectives were to: 1) estimate the diagnostic accuracy of cytology and culture of endometrial swab samples to detect endometritis in mares; 2) assess the impact of different cytology thresholds on test accuracy and misclassification costs; and 3) assess the sensitivity (Se) and specificity (Sp) of a diagnostic strategy including both tests interpreted in series and parallel. Diagnostic and pre-breeding endometrial swab samples collected from 3448 mares based at breeding premises located in the South East of England between 2014 and 2020 were retrospectively analysed. Culture results were classified as positive according to three different case definitions: (A) > 90% of the growth colonies were a monoculture; (B) pathogenic or pathogenic and non-pathogenic bacteria were identified; and (C) any growth was observed. Endometrial smears were graded based on the percent of polymorphonuclear cells (PMN) per high power field (HPF). A hierarchical BLCM was fitted using the cross-tabulated results of the three culture case definitions with a cytology threshold fixed at > 0.5% PMN. Fit for purpose cytology thresholds were proposed using a misclassification cost analysis in the context of good antimicrobial stewardship and for varying endometritis prevalence estimates. Median [95% Bayesian credible intervals (BCI)] cytology Se estimates were 6.5% (2.2–11.6), 6.4% (2.2–10.8) and 6.3% (2.2–10.8) for scenario A, B and C, respectively. Median (95% BCI) cytology Sp estimates were 88.8% (83.1–94.8), 88.9% (83.9–93.8) and 88.8% (84.0–93.8) for scenarios A, B and C, respectively. Median (95% BCI) culture Se estimates were 37.5% (29.9–46.0), 42.3% (33.8–51.1) and 46.4% (35.7–55.9) for scenarios A, B and C, respectively. Median (95% BCI) culture Sp estimates were 92.8% (84.3–99.0), 91.5% (82.5–98.0) and 90.8% (80.1–97.4) for scenarios A, B and C, respectively. Regardless of the culture case definition, Se and Sp of cytology (> 0.5% PMN) was lower than previously reported for swab samples in studies using histology as the reference standard test. The misclassification cost term decreased as the cytology threshold increased for all scenarios and all prevalence contexts, suggesting that, regardless of the endometritis prevalence in the population, increasing the cytology threshold would reduce the misclassification costs associated with false positive mares contributing to good antimicrobial stewardship.

Evaluation of 2 refractometers to estimate different passive immunity status in Simmental dairy calves

Inadequate transfer of passive immunity (ITPI) is major risk for mortality, morbidity and decreased growth performance in dairy and beef calves. Refractometry method is used in estimating ITPI in dairy and beef calves but studies evaluating refractometers for Simmental dairy calves are limited. The objectives of this study were determining the accuracy of digital serum total protein refractometer (STP-REF) and digital Brix refractometer (BRIX-REF) in the estimating of different passive immunity status (<10, <18, <25 and <32 g/L) in Simmental dairy calves. Serum samples were collected from apparently healthy Simmental dairy calves aged 1–8 days (n = 291). Serum total solid percentages were measured using BRIX-REF, serum total protein concentrations were measured using STP-REF and serum IgG concentrations were measured using radial immunodiffusion (RID) assay as a reference test. Correlation coefficients were calculated between RID test and refractometers results, and each other. The diagnostic test performance of the refractometers at IgG concentrations of < 10, < 18, < 25 and < 32 g/L was measured by performing a receiver operating characteristics (ROC) analysis. The optimal thresholds of the refractometers were determined based on Youden's J statistics and after accounting for different costs between false positive and false negative cases using misclassification cost term analysis. The overall test performance and the agreement of the refractometers were assessed using area under the ROC curve and Cohen's kappa analysis. Optimal thresholds were determined as < 7.9, < 8.3, < 8.7, < 9.4% for the BRIX-REF, and < 4.6, < 5.2, < 5.4, < 5.8 g/dL for the STP-REF in estimating IgG concentrations of < 10, < 18, < 25, < 32 g/L, respectively. Se and Sp of BRIX-REF were 88.8% and 89.1% for < 7.9% Brix, 81.6% and 94.2% for < 8.3% Brix, 77.9% and 97.4% for < 8.7% Brix and 81.7% and 91.2% for < 9.4% Brix, respectively. Se and Sp of STP-REF 92.5% and 88.2% for < 4.6 g/dL; 92.1% and 87.1% for < 5.2 g/dL; 81.7% and 93.6% for < 5.4 g/dL; 79.0% and 94.1% for < 5.8 g/dL, respectively. Both, digital Brix and STP refractometers performed well to estimate the status of passive immunity in dairy Simmental calves.

Development of a multivariable prediction model to identify dairy calves too young to be transported to auction markets in Canada using simple physical examination and body weight

Calves born on Eastern Canadian dairy farms that are not kept in the herds are traditionally sold through auction markets and are raised for meat purposes such as veal calves. Since February 2020, a new Canadian federal regulation has forbidden calves <9 d old to be sold through auction markets. However, in the absence of a real-time birth registry consultation system, it would be of interest to look for predictors that could be associated with age to allow identification of calves too young to be transported. In the current retrospective cross-sectional study, 1,178 calves with a declared birth date (411 calves aged <9 d old; 34.9%) were assessed in 2 large Québec auction sites. Easy-to-record covariates [body weight (BW), breed phenotype, and presence of an umbilical cord remnant] as well as other clinical signs (umbilical swelling, enlargement, umbilical pain, wet umbilicus, skin tent, sunken eyes, ocular and nasal secretion, and hide cleanliness) were assessed. Two logistic regression models using age as a dichotomous dependent variable (<9 d old vs ≥9 d old) were built. The first model (model 1) considered all covariates, which were selected after univariable analyses and a backward stepwise selection process, whereas a more pragmatic model (model 2) only included the 3 easy-to-record variables (i.e., BW, breed, umbilical cord). Both models had similar accuracy to detect calves <9 d old (sensitivity of 38.4 and 37.5%, and specificity of 85.7 and 84.6% for model 1 and 2, respectively). Model 2 was subsequently more specifically studied as it employs a faster and easier assessment. Decision thresholds were tested for their robustness based on misclassification cost term (MCT) analysis with various prevalence of calves <9 d old and various costs of false-negative:false-positive ratio. Despite statistical significance, model accuracy, even if refined with MCT analysis, was limited at the individual level, showing the limits of using physical signs and BW or their combination as a reliable proxy of age. The sensitivity of these models to find calves <9 d old was not to be used for monitoring compliance with the Canadian federal regulation. The relatively high model specificity may help to use this model as a rule-in test (i.e., targeting positive calves for further investigation) rather than a rule-out test (due to its low sensitivity).

Validation of serum gamma-glutamyl transferase activity and body weight information for identifying dairy calves that are too young to be transported to auction markets in Canada

Dairy calves are at risk of being stressed when transported during the first week of life. A new Canadian federal rule will forbid transportation of calves younger than 9 d old to auction market. However, in the absence of reliable information to determine birth date, other indirect methods would be of interest. This study aimed to determine the prediction accuracy of body weight, Brix refractometry, and serum gamma-glutamyl transferase (GGT) activity for determining if a calf was not fit to be transported (i.e., <9 d old). For this purpose, we used 284 calves with a known birth date from a cross-sectional and a prospective cohort study. A logistic regression model was built based on multivariable analysis as well as a misclassification cost term analysis. Because of the collinearity between GGT activity and Brix value and lower discrimination of Brix value, the GGT activity was retained for the main model. The final logistic regression model contained body weight and log-transformed GGT activity value. The misclassifications of the logistic model was minimized using a model probability threshold ≥0.55 with a sensitivity of 70.4% and a specificity of 77.3%. This probability threshold was relatively robust for various prevalence and false negative to false positive cost ratios. The prediction accuracy of this model was moderate at the individual level, but is helpful in calves with a reasonable suspicion of being less than 9 d old.

Mining corporate annual reports for intelligent detection of financial statement fraud – A comparative study of machine learning methods

Financial statement fraud has been serious concern for investors, audit firms, government regulators, and other capital market stakeholders. Intelligent financial statement fraud detection systems have therefore been developed to support decision-making of the stakeholders. Fraudulent misrepresentation of financial statements in managerial comments has been noticed in recent studies. As such, the purpose of this study was to examine whether an improved financial fraud detection system could be developed by combining specific features derived from financial information and managerial comments in corporate annual reports. To develop this system, we employed both intelligent feature selection and classification using a wide range of machine learning methods. We found that ensemble methods outperformed the remaining methods in terms of true positive rate (fraudulent firms correctly classified as fraudulent). In contrast, Bayesian belief networks (BBN) performed best on non-fraudulent firms (true negative rate). This finding is important because interpretable ``green flag” values (for which fraud is likely absent) could be derived, providing potential decision support to auditors during client selection or audit planning. We also observe that both financial statements and text in annual reports can be utilised to detect non-fraudulent firms. However, non-annual report data (analysts’ forecasts of revenues and earnings) are necessary to detect fraudulent firms. This finding has important implications for selecting variables when developing early warning systems of financial statement fraud.

Learning cost-sensitive Bayesian networks via direct and indirect methods

Cost-Sensitive learning has become an increasingly important area that recognizes that real world classification problems need to take the costs of misclassification and accuracy into account. Much work has been done on cost-sensitive decision tree learning, but very little has been done on cost-sensitive Bayesian networks. Although there has been significant research on Bayesian networks there has been relatively little research on learning cost-sensitive Bayesian networks. Hence, this paper explores whether it is possible to develop algorithms that learn cost-sensitive Bayesian networks by taking (i) an indirect approach that changes the data distribution to reflect the costs of misclassification; and (ii) a direct approach that amends an existing accuracy based algorithm for learning Bayesian networks. An empirical comparison of the new approaches is carried out with cost-sensitive decision tree learning algorithms on 33 data sets, and the results show that the new algorithms perform better in terms of misclassification cost and maintaining accuracy.

Prediction Models in Machine Learning by Classification and Regression

One of the machine-learning method for constructing prediction models from data is Classification and Regression. By partitioning the data space recursively these models are configuring and in each prediction model are fitting with a simple predictions. Finally, the partitioning can be represented pictorially as a decision tree. Finite number ofunordered values are taken for the designing the classification trees and are designed for independent variables. And the prediction error are measured in terms of misclassification cost. Squared difference between the predicted and observed values are measured in regression trees, which are dependent variables that have ordered discrete values or continuous values. Here in this article reviewing and comparing some of the widely acceptable algorithms such as QUEST, GUIDE, CRUISE, C4.5 and RPART with their strengths, weakness and capabilities.

English

The classification tree is commonly used in data mining for investigating interaction among predictors, particularly. The splitting rule and the decision trees technique employ algorithms that are largely based on statistical and probability methods. Splitting procedure is the most important phase of classification tree training. The aim of this study is to compare Gini and Twoing splitting rules in terms of misclassification cost, obtained the optimal balanced trees and the importance of independent variables. This study shows that the results obtained using the Twoing criterion, as it yields a tree that is much more equally balanced than the tree obtained with the Gini criterion. Misclassification rate was slightly different for the two methods (19% using Twoing criterion and 21,2% for the Gini).Using Twoing splitting rule gets more importance level independent variables and the improvement values are higher than the Gini algorithm. All things being considered, the good performance of the Twoing splitting in this study combined with its robustness to get high classification accuracy, tree structure and the importance of independent variables.

Prediction Models in Machine Learning by Classification and Regression

One of the machine-learning method for constructing prediction models from data is Classification and Regression. By partitioning the data space recursively these models are configuring and in each prediction model are fitting with a simple predictions. Finally, the partitioning can be represented pictorially as a decision tree. Finite number of unordered values are taken for the designing the classification trees and are designed for independent variables. And the prediction error are measured in terms of misclassification cost. Squared difference between the predicted and observed values are measured in regression trees, which are dependent variables that have ordered discrete values or continuous values. Here in this article reviewing and comparing some of the widely acceptable algorithms such as QUEST, GUIDE, CRUISE, C4.5 and RPART with their strengths, weakness and capabilities.

Assessment of l-lactatemia as a predictor of respiratory disease recognition and severity in feedlot steers

The bovine respiratory disease complex (BRD) is a major health issue in feedlot cattle and one of the primary reasons for antimicrobial use in the North American feedlot industry. The purpose of the present study was to assess blood l-lactate levels of feedlot steers at high risk of developing BRD during the early feeding period. Blood samples were obtained at initial processing and again after BRD confirmation (using bronchial lavage or thoracic ultrasound exam). The study involved 232 recently weaned steers received at a single research feedlot that were processed without metaphylactic antimicrobial treatment. Blood samples were obtained for determination of l-lactatemia and temperament scores (very quiet or stoic [score 1], average [score 2] and very excited [score 3]) were systematically assigned at initial processing. A subsample of calves that were later confirmed as cases of BRD were sampled at first pull (day 0), and at subsequent observation points on days 3, 6, 9 and 15 following initial BRD diagnosis for blood lactate determination as a potential indicator of subsequent death. The clinical BRD cumulative incidence in the cohort was 38% (87/232). Temperament was associated with the probability of becoming a BRD case during the early feeding period. Stoic or very excited calves showed 2.2 times higher odds (95%CI: 1.3, 3.8) of becoming BRD cases compared to calves with average temperament. The impact of l-lactatemia differed by temperament strata. In calves with a temperament score of 2 (average temperament) every 1-log unit increase of lactatemia at processing resulted in 1.9 times higher odds (95% CI: 1.2, 3.1) of becoming a BRD case; this relationship was not significant in calves with a score of either 1 or 3. Twenty-nine confirmed BRD cases were studied for the dynamic lactate assessment analysis. l-lactate at first pull was not significantly different between survivors (median 3.3mmol/L; range 0.8–7.8mmol/L) and non-survivors (median 2.7mmol/L; range: 1.6–5.4mmol/L) steers. However, the dynamic assessment of l-lactatemia was associated with the hazard of death using Cox proportional hazard survival analysis. A 1-log increase of lactatemia increased the hazard of dying prior to the next observation by a factor of 36.5 (95% CI: 3.5–381.6).For calves showing a normal temperament score (i.e. temperament score of 2), a misclassification cost term analysis was conducted to identify potential l-lactate test thresholds for identifying future BRD steers. When planned test usage was for informing decision of administering or not a metaphylactic treatment at processing, experts agreed that false-negative (not treating a calf that would have benefit from treatment) to false-positive (wrongfully treating a calf that would have remained healthy) health costs ratio ranged from 8:1 to 20:1. In this situation, a threshold of 5mmol/L would have best informed treatment decision. When using l-lactate for informing the type of antimicrobial used at processing, false-negative to false-positive health costs ratio ranging from 1:1 to 3:1 could be expected and, again, a l-lactate threshold of 5.0mmol/L would have minimized the costs associated with calves’ misclassification and could be used to identify calves that would benefit from a more efficient metaphylactic treatment. This study provides an interesting perspective on the potential application of chute-side markers or diagnostic tests to stratify the risk of future pull for BRD in cattle during processing in order to adapt antimicrobial treatments accordingly.

Preoperative cow-side lactatemia measurement predicts negative outcome in Holstein dairy cattle with right abomasal disorders

The objectives of the current study were (1) to determine the gain in prognostic accuracy of preoperative l-lactate concentration (LAC) measured on farm on cows with right displaced abomasum (RDA) or abomasal volvulus (AV) for predicting negative outcome; and (2) to suggest clinically relevant thresholds for such use. A cohort of 102 cows with on-farm surgical diagnostic of RDA or AV was obtained from June 2009 through December 2011. Blood was drawn from coccygeal vessels before surgery and plasma LAC was immediately measured by using a portable clinical analyzer. Dairy producers were interviewed by phone 30d following surgery and the outcome was determined: a positive outcome if the owner was satisfied of the overall evolution 30d postoperatively, and a negative outcome if the cow was culled, died, or if the owner reported being unsatisfied 30d postoperatively. The area under the curve of the receiver operating characteristic curve for LAC was 0.92 and was significantly greater than the area under the curve of the receiver operating characteristic curve of heart rate (HR; 0.77), indicating that LAC, in general, performed better than HR to predict a negative outcome. Furthermore, the ability to predict a negative outcome was significantly improved when LAC measurement was considered in addition to the already available HR data (area under the curve: 0.93 and 95% confidence interval: 0.87, 0.99). Important inflection points of the misclassification cost term function were noted at thresholds of 2 and 6mmol/L, suggesting the potential utility of these cut-points. The 2 and 6mmol/L thresholds had a sensitivity, specificity, positive predictive value, and negative predictive value for predicting a negative outcome of 76.2, 82.7, 53.3, and 93.1%, and of 28.6, 97.5, 75, and 84%, respectively. In terms of clinical interpretation, LAC ≤2mmol/L appeared to be a good indicator of positive outcome and could be used to support a surgical treatment decision. The treatment decision for cows with LAC between 2 and 6mmol/L, however, would depend on the economic context and the owner’s attitude to risk in regard to potential return on its investment. Finally, performing a surgical correction on commercial cows with RDA or AV and a LAC ≥6mmol/L appeared to be unjustified and these animals should be culled based on their high probability of negative outcome.

Classification and regression trees

AbstractClassification and regression trees are machine‐learning methods for constructing prediction models from data. The models are obtained by recursively partitioning the data space and fitting a simple prediction model within each partition. As a result, the partitioning can be represented graphically as a decision tree. Classification trees are designed for dependent variables that take a finite number of unordered values, with prediction error measured in terms of misclassification cost. Regression trees are for dependent variables that take continuous or ordered discrete values, with prediction error typically measured by the squared difference between the observed and predicted values. This article gives an introduction to the subject by reviewing some widely available algorithms and comparing their capabilities, strengths, and weakness in two examples. © 2011 John Wiley &amp; Sons, Inc. WIREs Data Mining Knowl Discov 2011 1 14‐23 DOI: 10.1002/widm.8This article is categorized under: Technologies &gt; Classification Technologies &gt; Machine Learning Technologies &gt; Prediction Algorithmic Development &gt; Statistics

A neuro-genetic based short-term forecasting framework for network intrusion prediction system

Information systems are one of the most rapidly changing and vulnerable systems, where security is a major issue. The number of security-breaking attempts originating inside organizations is increasing steadily. Attacks made in this way, usually done by “authorized” users of the system, cannot be immediately traced. Because the idea of filtering the traffic at the entrance door, by using firewalls and the like, is not completely successful, the use of intrusion detection systems should be considered to increase the defense capacity of an information system. An intrusion detection system (IDS) is usually working in a dynamically changing environment, which forces continuous tuning of the intrusion detection model, in order to maintain sufficient performance. The manual tuning process required by current IDS depends on the system operators in working out the tuning solution and in integrating it into the detection model. Furthermore, an extensive effort is required to tackle the newly evolving attacks and a deep study is necessary to categorize it into the respective classes. To reduce this dependence, an automatically evolving anomaly IDS using neuro-genetic algorithm is presented. The proposed system automatically tunes the detection model on the fly according to the feedback provided by the system operator when false predictions are encountered. The system has been evaluated using the Knowledge Discovery in Databases Conference (KDD 2009) intrusion detection dataset. Genetic paradigm is employed to choose the predominant features, which reveal the occurrence of intrusions. The neuro-genetic IDS (NGIDS) involves calculation of weightage value for each of the categorical attributes so that data of uniform representation can be processed by the neuro-genetic algorithm. In this system unauthorized invasion of a user are identified and newer types of attacks are sensed and classified respectively by the neuro-genetic algorithm. The experimental results obtained in this work show that the system achieves improvement in terms of misclassification cost when compared with conventional IDS. The results of the experiments show that this system can be deployed based on a real network or database environment for effective prediction of both normal attacks and new attacks.

Measuring classifier performance: a coherent alternative to the area under the ROC curve

The area under the ROC curve (AUC) is a very widely used measure of performance for classification and diagnostic rules. It has the appealing property of being objective, requiring no subjective input from the user. On the other hand, the AUC has disadvantages, some of which are well known. For example, the AUC can give potentially misleading results if ROC curves cross. However, the AUC also has a much more serious deficiency, and one which appears not to have been previously recognised. This is that it is fundamentally incoherent in terms of misclassification costs: the AUC uses different misclassification cost distributions for different classifiers. This means that using the AUC is equivalent to using different metrics to evaluate different classification rules. It is equivalent to saying that, using one classifier, misclassifying a class 1 point is p times as serious as misclassifying a class 0 point, but, using another classifier, misclassifying a class 1 point is P times as serious, where p≠P. This is nonsensical because the relative severities of different kinds of misclassifications of individual points is a property of the problem, not the classifiers which happen to have been chosen. This property is explored in detail, and a simple valid alternative to the AUC is proposed.

Interethnic differences in the accuracy of anthropometric indicators of obesity in screening for high risk of coronary heart disease.

Cut points for defining obesity have been derived from mortality data among Whites from Europe and the United States and their accuracy to screen for high risk of coronary heart disease (CHD) in other ethnic groups has been questioned. To compare the accuracy and to define ethnic and gender-specific optimal cut points for body mass index (BMI), waist circumference (WC) and waist-to-hip ratio (WHR) when they are used in screening for high risk of CHD in the Latin-American and the US populations. We estimated the accuracy and optimal cut points for BMI, WC and WHR to screen for CHD risk in Latin Americans (n=18 976), non-Hispanic Whites (Whites; n=8956), non-Hispanic Blacks (Blacks; n=5205) and Hispanics (n=5803). High risk of CHD was defined as a 10-year risk > or =20% (Framingham equation). The area under the receiver operator characteristic curve (AUC) and the misclassification-cost term were used to assess accuracy and to identify optimal cut points. WHR had the highest AUC in all ethnic groups (from 0.75 to 0.82) and BMI had the lowest (from 0.50 to 0.59). Optimal cut point for BMI was similar across ethnic/gender groups (27 kg/m(2)). In women, cut points for WC (94 cm) and WHR (0.91) were consistent by ethnicity. In men, cut points for WC and WHR varied significantly with ethnicity: from 91 cm in Latin Americans to 102 cm in Whites, and from 0.94 in Latin Americans to 0.99 in Hispanics, respectively. WHR is the most accurate anthropometric indicator to screen for high risk of CHD, whereas BMI is almost uninformative. The same BMI cut point should be used in all men and women. Unique cut points for WC and WHR should be used in all women, but ethnic-specific cut points seem warranted among men.

An evaluation of alternative scoring models in private banking

PurposeThis paper aims to investigate the efficiency and effectiveness of alternative credit‐scoring models for consumer loans in the banking sector. In particular, the focus is upon the financial risks associated with both the efficiency of alternative models in terms of correct classification rates, and their effectiveness in terms of misclassification costs (MCs).Design/methodology/approachA data set of 630 loan applicants was provided by an Egyptian private bank. A two‐thirds training sample was selected for building the proposed models, leaving a one‐third testing sample to evaluate the predictive ability of the models. In this paper, an investigation is conducted into both neural nets (NNs), such as probabilistic and multi‐layer feed‐forward neural nets, and conventional techniques, such as the weight of evidence measure, discriminant analysis and logistic regression.FindingsThe results revealed that a best net search, which selected a multi‐layer feed‐forward net with five nodes, generated both the most efficient classification rate and the most effective MC. In general, NNs gave better average correct classification rates and lower MCs than traditional techniques.Practical implicationsBy reducing the financial risks associated with loan defaults, banks can achieve a more effective management of such a crucial component of their operations, namely, the provision of consumer loans.Originality/valueThe use of NNs and conventional techniques in evaluating consumer loans within the Egyptian private banking sector utilizes rigorous techniques in an environment which merits investigation.

An Automatically Tuning Intrusion Detection System

An intrusion detection system (IDS) is a security layer used to detect ongoing intrusive activities in information systems. Traditionally, intrusion detection relies on extensive knowledge of security experts, in particular, on their familiarity with the computer system to be protected. To reduce this dependence, various data-mining and machine learning techniques have been deployed for intrusion detection. An IDS is usually working in a dynamically changing environment, which forces continuous tuning of the intrusion detection model, in order to maintain sufficient performance. The manual tuning process required by current systems depends on the system operators in working out the tuning solution and in integrating it into the detection model. In this paper, an automatically tuning IDS (ATIDS) is presented. The proposed system will automatically tune the detection model on-the-fly according to the feedback provided by the system operator when false predictions are encountered. The system is evaluated using the KDDCup'99 intrusion detection dataset. Experimental results show that the system achieves up to 35% improvement in terms of misclassification cost when compared with a system lacking the tuning feature. If only 10% false predictions are used to tune the model, the system still achieves about 30% improvement. Moreover, when tuning is not delayed too long, the system can achieve about 20% improvement, with only 1.3% of the false predictions used to tune the model. The results of the experiments show that a practical system can be built based on ATIDS: system operators can focus on verification of predictions with low confidence, as only those predictions determined to be false will be used to tune the detection model.

Bayesian estimation of sensitivity and specificity of serum ELISA and faecal culture for diagnosis of paratuberculosis in Greek dairy sheep and goats

Latent class models were used to estimate the sensitivity (Se) and the specificity (Sp) of a serum ELISA and a faecal culture (FC) method for the diagnosis of paratuberculosis separately, in sheep and goats. The estimates were obtained by a Bayesian method. Possible dependence of diagnostic errors was investigated by comparing models where independence was assumed to models allowing for conditional dependence given the true disease status. ROC analysis for the serum ELISA was also performed and optimized cut-off values based on the misclassification cost term were determined. No evidence of conditional dependence was found. Assuming independence, posterior medians and 95% credible intervals for the Se ELISA, Sp ELISA, Se FC and Sp FC, were 63% (42, 93%), 95% (90, 98%), 8% (2, 17%) and 98% (95, 100%) in goats and 37% (10, 80%), 97% (93, 99%), 16% (2, 48%) and 97% (95, 99%) in sheep. AUC was calculated 0.702 for sheep and 0.847 for goats. For the serum ELISA, there is need of species- and purpose-specific cut-off selection. For instance, with 20% prevalence situation and assuming equal and five-fold cost of a false negative to a false positive test result, the optimal cut-off is 0.3 and 0.05 in sheep, respectively, while it is 0.6 and 0.1 in goats, respectively. Serum ELISA performed better in goats than in sheep. Lowering the cut-off, in relation to the one recommended by the manufacturer, improved Se ELISA without seriously compromising Sp ELISA, in either species.

Validation of IgG-sandwich and IgM-capture ELISA for the detection of antibody to Rift Valley fever virus in humans

Rift Valley fever (RVF) virus is an important zoonotic and a potential biothreat agent. This paper describes validation of sandwich and capture enzyme-linked immunoassays (ELISA) based on gamma-irradiated antigens for the detection of RVFV-specific IgG and IgM antibody in humans. Validation data sets derived from testing field-collected sera from Africa ( n = 2400) were dichotomised according to the results of a virus neutralisation test. In addition, sera from laboratory workers immunized with inactivated RVF vaccine ( n = 93) and serial sera ( n = 3) from a single RVF case were used. ELISA data were expressed as percentage of high-positive control serum (PP). Cut-off values at 95% accuracy level were optimised using the misclassification cost term option of the two-graph receiver operating characteristics analysis. During the routine use of assays there was no evidence for excessive intra- and inter-plate variations within and between runs of assays. At a cut-off of 13.2 PP the sensitivity of the IgG-sandwich ELISA was 100% and specificity 99.95%, while for the IgM-capture ELISA the values were 96.47 and 99.44%, respectively, at a cut-off of 7.1 PP. Compared to the virus neutralisation test, the IgG-sandwich ELISA was more sensitive in detection of immunological responses in vaccines. Following natural infection class-specific antibodies were detected in serum taken 6 days after onset of symptoms. The results demonstrate that both assays will be useful for early diagnosis of infection, epidemiological surveillance and for monitoring of immune response after vaccination. As highly accurate, robust and safe tests, they have the potential to replace traditional diagnostic methods which are unable to distinguish between different classes of immunoglobulins, and pose health risks necessitating their use being restricted to high containment facilities outside RVF endemic areas.