The potential of neural networks for classification problems has been established by numerous successful applications reported in the literature. One of the major assumptions used in almost all studies is the equal cost consequence of misclassification. With this assumption, minimizing the total number of misclassification errors is the sole objective in developing a neural network classifier. Often this is done simply to ease model development and the selection of classification decision points. However, it is not appropriate for many real situations such as quality assurance, direct marketing, bankruptcy prediction, and medical diagnosis where misclassification costs have unequal consequences for different categories. In this paper, we investigate the issue of unequal misclassification costs in neural network classifiers. Through an application in thyroid disease diagnosis, we find that different cost considerations have significant effects on the classification performance and that appropriate use of cost information can aid in optimal decision making. A cross‐validation technique is employed to alleviate the problem of bias in the training set and to examine the robustness of neural network classifiers with regard to sampling variations and cost differences.
AbstractIf misclassification occurs the standard binomial estimator is usually seriously biased. It is known that an improvement can be achieved by using more than one observer in classifying the sample elements. Here it will be investigated which number of observers is optimal given the total number of judgements that can be made. An adaptive estimator for the probability of interest is introduced which uses an estimator of this optimal number of observers, obtained without additional cost. Some simulation results are presented which suggest that the adaptive procedure performs quite well.
This brief examines the history regulation and organizing short-haul trucking at the ports of Los Angeles and Long Beach. During a long period of high union density and strong regulation, short haul truck-driving grew into a desirable blue-collar occupation. However, deregulation during the 1970s and 1980s incentivized trucking companies to shift their workforces to owner-operator truckers in response to competitive pressures. Amplified by rampant misclassification, this arrangement shifts risks and costs onto drivers while exempting them from labor protections and the ability to form a union. Wage theft in the form of lengthy and uncompensated wait times is also common. The brief concludes with a consideration of the applicability of model legislation from other states to the particularly acute issues faced by California's short-haul truckers.
Worker misclassification is a form of precarious employment in which employers illegally designate their employees as 'independent contractors' to cut labor costs. Non-standard employment arrangements and the emergence of the misclassification problem are expressions of neoliberal economic reform and attendant shifts in managerial strategy. Although scholars and government statisticians have documented the prevalence of worker misclassification, extant research on labor-organizing campaigns in response to this practice is lacking. This paper presents case studies of two successful organizing campaigns against worker misclassification: (1) a United Brotherhood of Carpenters and Joiners of America (UBCJA) effort in the Northeastern construction industry and (2) a Teamsters campaign focused on the West Coast port trucking industry. Both campaigns employ similar frames highlighting competition, free markets, and the necessity of industrial change to achieve these ideals. We conclude with a discussion of the prospects and limitations of these organizing strategies given the countervailing political and economic headwinds posed by neoliberal restructuring.
Poster Division: Arts, Humanities, and Social Sciences: 3rd Place (The Ohio State University Edward F. Hayes Graduate Research Forum) ; A fundamental challenge for state Medicaid programs is the ongoing task of implementing beneficiary enrollment according to specified eligibility criteria. Errors in enrollment present themselves in the form of eligible individuals not taking up coverage (missed take-up), individuals continually moving on and off the program (churn), and ineligible individuals receiving coverage (fraud). In addition to posing problems for the coverage and continuity of care of vulnerable populations, enrollment errors create difficulties for state program planning, budgeting, and contract arrangements. Despite passage of the Affordable Care Act in 2010, state Medicaid programs remain varied in the structure of their eligibility categories and their processes for determination and enrollment of beneficiaries. There are multiple levels of prescribed action and discretion in these systems, and both short (month-to-month) and long (multiple year) time horizons are of interest. Previous studies model enrollment outcomes as a deterministic function of personal characteristics (e.g., race/ethnicity) and general economic indicators (e.g., unemployment rate). The policies and procedures defining design and administration of the program are rarely included in models estimating enrollment patterns, thus ignoring potentially important sources of enrollment dynamics. The purpose of this research is to understand the mechanism by which Medicaid eligibility criteria are transformed into enrollment outcomes among state program populations. Drawing on insights from systems science and implementation research, this study uses simulated experiments with program structure to describe and explain the dynamics of Medicaid eligibility determination and enrollment processes. Program-level state Medicaid enrollment patterns are modeled in terms of the dynamics among individual characteristics, program eligibility criteria, and administrative procedures within relevant social, economic, demographic, and political contexts. The effects of federal guidelines, along with exemplary cases of state eligibility rules and determination procedures, are analyzed through a set of system dynamics models of program-level enrollment patterns. This study tests the possible effects of a number of implementation strategies drawn from the Medicaid enrollment literature on eligibility determination and benefit enrollment errors. Interrelationships among defined eligibility categories and (re)determination procedures create enrollment implementation error in Medicaid systems even when household economic conditions and decision making are stable. Delays in (re)determination processes lead to accumulations of households in enrollment states not consistent with their eligibility. A nonlinear relationship between the demand for benefits and the administrative capacity to process applications leads to further accumulation of misclassified households. Enrollment errors, including missed take-up, churn, and fraud, are shown to arise form the structure and dynamics of the program system itself, rather than solely from individual circumstance or exogenous economic shock. These simulations allow policy makers and scholars to experiment with various implementation strategies with effectively zero social costs. Illuminating the core dynamics of the Medicaid enrollment mechanism aids in forecasting enrollment and spending levels, revealing administrative leverage points to improve system performance, and evaluating potential tradeoffs between costs and coverage over time. Additionally, the simulations can be tailored to specific economic, demographic, and program conditions, providing a picture of the possible range of outcomes associated with specific administrative actions. ; No embargo
The question of how to obtain a specified quantity of satisfactory manpower inputs for a given time period is a problem which confronts almost any personnel decisionmaker. The staffing model proposed in this paper defines selection, training, and any selection/training combination as being alternative staffing strategies which are available for coping with an organization's manpower requirement problems. Very few personnel decisionmakers, however, approach such problems with any appreciable knowledge of the relative costs associated with different alternative input strategies and of the costs of different misclassification errors which can result. The staffing model described here incorporates cost and utility considerations and suggests that the relative costs of alternative staffing strategies are amenable to systematic analysis. The model specifies the probability estimates for success and failure in both selection and training, the direct costs for each, and the potential costs which result from errors of misclassification. Then, an expression and rationale for determining the minimum total cost associated with various combinations of selection and training strategies are presented.
This book proposes complex hierarchical deep architectures (HDA) for predicting bankruptcy, a topical issue for business and corporate institutions that in the past has been tackled using statistical, market-based and machine-intelligence prediction models. The HDA are formed through fuzzy rough tensor deep staking networks (FRTDSN) with structured, hierarchical rough Bayesian (HRB) models. FRTDSN is formalized through TDSN and fuzzy rough sets, and HRB is formed by incorporating probabilistic rough sets in structured hierarchical Bayesian model. Then FRTDSN is integrated with HRB to form the compound FRTDSN-HRB model. HRB enhances the prediction accuracy of FRTDSN-HRB model. The experimental datasets are adopted from Korean construction companies and American and European non-financial companies, and the research presented focuses on the impact of choice of cut-off points, sampling procedures and business cycle on the accuracy of bankruptcy prediction models. The book also highlights the fact that misclassification can result in erroneous predictions leading to prohibitive costs to investors and the economy, and shows that choice of cut-off point and sampling procedures affect rankings of various models. It also suggests that empirical cut-off points estimated from training samples result in the lowest misclassification costs for all the models. The book confirms that FRTDSN-HRB achieves superior performance compared to other statistical and soft-computing models. The experimental results are given in terms of several important statistical parameters revolving different business cycles and sub-cycles for the datasets considered and are of immense benefit to researchers working in this area.
AbstractIntroductionStrategies employing a single rapid diagnostic test (RDT) such as HIV self‐testing (HIVST) or "test for triage" (T4T) are proposed to increase HIV testing programme impact. Current guidelines recommend serial testing with two or three RDTs for HIV diagnosis, followed by retesting with the same algorithm to verify HIV‐positive status before anti‐retroviral therapy (ART) initiation. We investigated whether clients presenting to HIV testing services (HTS) following a single reactive RDT must undergo the diagnostic algorithm twice to diagnose and verify HIV‐positive status, or whether a diagnosis with the setting‐specific algorithm is adequate for ART initiation.MethodsWe calculated (1) expected number of false‐positive (FP) misclassifications per 10,000 HIV negative persons tested, (2) positive predictive value (PPV) of the overall HIV testing strategy compared to the WHO recommended PPV ≥99%, and (3) expected cost per FP misclassified person identified by additional verification testing in a typical low‐/middle‐income setting, compared to the expected lifetime ART cost of $3000. Scenarios considered were as follows: 10% prevalence using two serial RDTs for diagnosis, 1% prevalence using three serial RDTs, and calibration using programmatic data from Malawi in 2017 where the proportion of people testing HIV positive in facilities was 4%.ResultsIn the 10% HIV prevalence setting with a triage test, the expected number of FP misclassifications was 0.86 per 10,000 tested without verification testing and the PPV was 99.9%. In the 1% prevalence setting, expected FP misclassifications were 0.19 with 99.8% PPV, and in the Malawi 2017 calibrated setting the expected misclassifications were 0.08 with 99.98% PPV. The cost per FP identified by verification testing was $5879, $3770, and $24,259 respectively. Results were sensitive to assumptions about accuracy of self‐reported reactive results and whether reactive triage test results influenced biased interpretation of subsequent RDT results by the HTS provider.ConclusionsDiagnosis with the full algorithm following presentation with a reactive triage test is expected to achieve PPV above the 99% threshold. Continuing verification testing prior to ART initiation remains recommended, but HIV testing strategies involving HIVST and T4T may provide opportunities to maintain quality while increasing efficiency as part of broader restructuring of HIV testing service delivery.
Tax evasion refers to an entity indulging in illegal activities to avoid paying their actual tax liability. A tax return statement is a periodic report comprising information about income, expenditure, etc. One of the most basic tax evasion methods is failing to file tax returns or delay filing tax return statements. The taxpayers who do not file their returns, or fail to do so within the stipulated period are called tax return defaulters. As a result, the Government has to bear the financial losses due to a taxpayer defaulting, which varies for each taxpayer. Therefore, while designing any statistical model to predict potential return defaulters, we have to consider the real financial loss associated with the misclassification of each individual. This paper proposes a framework for an example-dependent cost-sensitive stacking classifier that uses cost-insensitive classifiers as base generalizers to make predictions on the input space. These predictions are used to train an example-dependent cost-sensitive meta generalizer. Based on the meta-generalizer choice, we propose four variant models used to predict potential return defaulters for the upcoming tax-filing period. These models have been developed for the Commercial Taxes Department, Government of Telangana, India. Applying our proposed variant models to GST data, we observe a significant increase in savings compared to conventional classifiers. Additionally, we develop an empirical study showing that our approach is more adept at identifying potential tax return defaulters than existing example-dependent cost-sensitive classification algorithms.
