Bayesian model averaging of load demand forecasts from neural network models

Creating a set of a number of neural network (NN) models in an ensemble and accumulating them can achieve better overview capability as compared to single neural network. Neural network ensembles are designed to provide solutions to particular problems. Many researchers and academicians have adopted this NN ensemble technique, especially in machine learning, and has been applied in various fields of engineering, medicine and information technology. This paper present a robust aggregation methodology for load demand forecasting based on Bayesian Model Averaging of a set of neural network models in an ensemble. This paper estimate a vector of coefficient for individual NN models' forecasts using validation data-set. These coefficients, also known as weights, are equal to posterior probabilities of the models generating the forecasts. These BMA weights are then used in combining forecasts generated from NN models with test data-set. By comparing the Bayesian results with the Simple Averaging method, it was observed that benefits are obtained by utilizing an advanced method like BMA for forecast combinations.

Entecavir for the treatment of chronic hepatitis B infection

This paper presents a summary of the evidence review group (ERG) report into the clinical and cost-effectiveness of entecavir for the treatment of chronic hepatitis B (CHB) in adults based upon a review of the manufacturer's submission to the National Institute for Health and Clinical Excellence (NICE) as part of the single technology appraisal (STA) process. The submission's evidence came from five randomised controlled trials (RCTs), of good methodological quality and measuring a range of clinically relevant outcomes, comparing entecavir with lamivudine. After 1 year of treatment entecavir was statistically superior to lamivudine in terms of the proportion of patients achieving hepatitis B virus (HBV) DNA suppression, alanine aminotransferase (ALT) normalisation and histological improvement, but not in terms of the proportion of patients achieving hepatitis B e antigen (HBeAg) seroconversion. The incidence of adverse or serious adverse events was similar for both treatments. The results of the manufacturer's mixed treatment comparison (MTC) model to compare entecavir with the comparator drugs in nucleoside-naive patients were considered to be uncertain because of concerns over its conduct and reporting. For the economic evaluation the manufacturer constructed two Markov state transition models, one in HBeAg-positive and one in HBeAg-negative patients. The modelling approach was considered reasonable subject to some uncertainties and concerns over some of the structural assumptions. In HBeAg-positive patients the base-case incremental cost-effectiveness ratios (ICER) for entecavir compared with lamivudine and pegylated interferon alpha-2a were 14,329 pounds and 8403 pounds per quality-adjusted life-year (QALY) respectively. Entecavir was dominated by telbivudine. In HBeAg-negative patients the base-case ICERs for entecavir compared with lamivudine, pegylated interferon alpha-2a and telbivudine were 13,208 pounds, 7511 pounds and 6907 pounds per QALY respectively. In HBeAg-positive lamivudine-refractory patients entecavir dominated adefovir added to lamivudine. In one-way deterministic sensitivity analysis on all key input parameters for entecavir compared with lamivudine in nucleoside-naive patients, ICERs generally remained under 30,000 pounds per QALY. In probabilistic sensitivity analysis in nucleoside-naive HBeAg-positive patients the probability of the ICER for entecavir being below 20,000 pounds per QALY was 57%, 82% and 45% compared with lamivudine, pegylated interferon alpha-2a and telbivudine respectively. In nucleoside-naive HBeAg-negative patients the probabilities were 90%, 100% and 96% respectively. The manufacturer's lifetime treatment scenario for HBeAg-negative patients and the ERG's 20-year treatment scenario for HBeAg-positive patients increased the ICERs, particularly in the latter case. Amending the HBeAg-negative model so that patients with compensated cirrhosis would also receive lifetime treatment gave probabilities of entecavir being cost-effective at a willingness to pay of 20,000 pounds and 30,000 pounds of 4% and 40% respectively. The NICE guidance issued in August 2008 as a result of the STA states that entecavir is recommended as an option for the treatment of people with chronic HBeAg-positive or HBeAg-negative hepatitis B in whom antiviral treatment is indicated.

Disease mapping and regression with count data in the presence of overdispersion and spatial autocorrelation: a Bayesian model averaging approach

This paper applies the generalised linear model for modelling geographical variation to esophageal cancer incidence data in the Caspian region of Iran. The data have a complex and hierarchical structure that makes them suitable for hierarchical analysis using Bayesian techniques, but with care required to deal with problems arising from counts of events observed in small geographical areas when overdispersion and residual spatial autocorrelation are present. These considerations lead to nine regression models derived from using three probability distributions for count data: Poisson, generalised Poisson and negative binomial, and three different autocorrelation structures. We employ the framework of Bayesian variable selection and a Gibbs sampling based technique to identify significant cancer risk factors. The framework deals with situations where the number of possible models based on different combinations of candidate explanatory variables is large enough such that calculation of posterior probabilities for all models is difficult or infeasible. The evidence from applying the modelling methodology suggests that modelling strategies based on the use of generalised Poisson and negative binomial with spatial autocorrelation work well and provide a robust basis for inference.

Managing B2B customer churn, retention and profitability

It is now widely accepted that firms should direct more effort into retaining existing customers than to attracting new ones. To achieve this, customers likely to defect need to be identified so that they can be approached with tailored incentives or other bespoke retention offers. Such strategies call for predictive models capable of identifying customers with higher probabilities of defecting in the relatively near future. A review of the extant literature on customer churn models reveals that although several predictive models have been developed to model churn in B2C contexts, the B2B context in general, and non-contractual settings in particular, have received less attention in this regard. Therefore, to address these gaps, this study proposes a data-mining approach to model non-contractual customer churn in B2B contexts. Several modeling techniques are compared in terms of their ability to predict true churners. The best performing data-mining technique (boosting) is then applied to develop a profit maximizing retention campaign. Results confirm that the model driven approach to churn prediction and developing retention strategies outperforms commonly used managerial heuristics. © 2014 Elsevier Inc.

Projected age- and sex-specific prevalence of cardiovascular diseases in Western Australian adults from 2005-2045

For decades, the incidence and mortality of cardiovascular diseases (CVDs) have declined. More recently, we have seen a halting in these declines, especially at younger ages. It is difficult to predict how these changing trends will impact CVD prevalence. We aimed to predict future prevalence of CVDs in Western Australian adults from 2005-2045 based on current incidence and mortality probabilities, population growth and ageing.

Estimating cumulative exposure of wildlife to non-lethal disturbance using spatially explicit capture-recapture models

Impact assessments often focus on short-term behavioral responses of animals to human disturbance. However, the cumulative effects caused by repeated behavioral disruptions are of management concern because these effects have the potential to influence individuals' survival and reproduction. We need to estimate individual exposure rates to disturbance to determine cumulative effects. We present a new approach to estimate the spatial exposure of minke whales to whalewatching boats in Faxaflõi Bay, Iceland. We used recent advances in spatially explicit capture-recapture modeling to estimate the probability that whales would encounter a disturbance (i.e., whalewatching boat). We obtained spatially explicit individual encounter histories of individually identifiable animals using photo-identification. We divided the study area into 1-km2 grid cells and considered each cell a spatially distinct sampling unit. We used capture history of individuals to model and estimate spatial encounter probabilities of individual minke whales across the study area, accounting for heterogeneity in sampling effort. We inferred the exposure of individual minke whales to whalewatching vessels throughout the feeding season by estimating individual whale encounters with vessels using the whale encounter probabilities and spatially explicit whalewatching intensity in the same area, obtained from recorded whalewatching vessel tracks. We then estimated the cumulative time whales spent with whalewatching boats to assess the biological significance of whalewatching disturbances. The estimated exposure levels to boats varied considerably between individuals because of both temporal and spatial variations in the activity centers of whales and the whalewatching intensity in the area. However, although some whales were repeatedly exposed to whalewatching boats throughout the feeding season, the estimated cumulative time they spent with boats was very low. Although whalewatching boat interactions caused feeding disruptions for the whales, the estimated low cumulative exposure indicated that the whalewatching industry in its current state likely is not having any long-term negative effects on vital rates.

Is there an interaction between socioeconomic status and FRAX 10-year fracture probability determined with and without bone density measures? Data from the Geelong Osteoporosis Study of female cohort.

FRAX(©) evaluates 10-year fracture probabilities and can be calculated with and without bone mineral density (BMD). Low socioeconomic status (SES) may affect BMD, and is associated with increased fracture risk. Clinical risk factors differ by SES; however, it is unknown whether aninteraction exists between SES and FRAX determined with and without the BMD. From the Geelong Osteoporosis Study, we drew 819 females aged ≥50 years. Clinical data were collected during 1993-1997. SES was determined by cross-referencing residential addresses with Australian Bureau of Statistics census data and categorized in quintiles. BMD was measured by dual energy X-ray absorptiometry at the same time as other clinical data were collected. Ten-year fracture probabilities were calculated using FRAX (Australia). Using multivariable regression analyses, we examined whether interactions existed between SES and 10-year probability for hip and any major osteoporotic fracture (MOF) defined by use of FRAX with and without BMD. We observed a trend for a SES * FRAX(no-BMD) interaction term for 10-year hip fracture probability (p = 0.09); however, not for MOF (p = 0.42). In women without prior fracture (n = 518), we observed a significant SES * FRAX(no-BMD) interaction term for hip fracture (p = 0.03) and MOF (p = 0.04). SES does not appear to have an interaction with 10-year fracture probabilities determined by FRAX with and without BMD in women with previous fracture; however, it does appear to exist for those without previous fracture.

A noisy-smoothing relevance feedback method for content-based medical image retrieval

In this paper, we address a new problem of noisy images which present in the procedure of relevance feedback for medical image retrieval. We concentrate on the noisy images, caused by the users mislabeling some irrelevant images as relevant ones, and a noisy-smoothing relevance feedback (NS-RF) method is proposed. In NS-RF, a two-step strategy is proposed to handle the noisy images. In step 1, a noisy elimination algorithm is adopted to identify and eliminate the noisy images. In step 2, to further alleviate the influence of noisy images, a fuzzy membership function is employed to estimate the relevance probabilities of retained relevant images. After noisy handling, the fuzzy support vector machine, which can take into account different relevant images with different relevance probabilities, is adopted to re-rank the images. The experimental results on the IRMA medical image collection demonstrate that the proposed method can deal with the noisy images effectively.

Visualization of HIV-1 interactions with penile and foreskin epithelia: clues for female-to-male HIV transmission

To gain insight into female-to-male HIV sexual transmission and how male circumcision protects against this mode of transmission, we visualized HIV-1 interactions with foreskin and penile tissues in ex vivo tissue culture and in vivo rhesus macaque models utilizing epifluorescent microscopy. 12 foreskin and 14 cadaveric penile specimens were cultured with R5-tropic photoactivatable (PA)-GFP HIV-1 for 4 or 24 hours. Tissue cryosections were immunofluorescently imaged for epithelial and immune cell markers. Images were analyzed for total virions, proportion of penetrators, depth of virion penetration, as well as immune cell counts and depths in the tissue. We visualized individual PA virions breaching penile epithelial surfaces in the explant and macaque model. Using kernel density estimated probabilities of localizing a virion or immune cell at certain tissue depths revealed that interactions between virions and cells were more likely to occur in the inner foreskin or glans penis (from local or cadaveric donors, respectively). Using statistical models to account for repeated measures and zero-inflated datasets, we found no difference in total virions visualized at 4 hours between inner and outer foreskins from local donors. At 24 hours, there were more virions in inner as compared to outer foreskin (0.0495 +/- 0.0154 and 0.0171 +/- 0.0038 virions/image, p = 0.001). In the cadaveric specimens, we observed more virions in inner foreskin (0.0507 +/- 0.0079 virions/image) than glans tissue (0.0167 +/- 0.0033 virions/image, p<0.001), but a greater proportion was seen penetrating uncircumcised glans tissue (0.0458 +/- 0.0188 vs. 0.0151 +/- 0.0100 virions/image, p = 0.099) and to significantly greater mean depths (29.162 +/- 3.908 vs. 12.466 +/- 2.985 μm). Our in vivo macaque model confirmed that virions can breach penile squamous epithelia in a living model. In summary, these results suggest that the inner foreskin and glans epithelia may be important sites for HIV transmission in uncircumcised men.

An integrated framework for suicide risk prediction

Suicide is a major concern in society. Despite of great attention paid by the community with very substantive medico-legal implications, there has been no satisfying method that can reliably predict the future attempted or completed suicide. We present an integrated machine learning framework to tackle this challenge. Our proposed framework consists of a novel feature extraction scheme, an embedded feature selection process, a set of risk classifiers and finally, a risk calibration procedure. For temporal feature extraction, we cast the patient’s clinical history into a temporal image to which a bank of one-side filters are applied. The responses are then partly transformed into mid-level features and then selected in 1-norm framework under the extreme value theory. A set of probabilistic ordinal risk classifiers are then applied to compute the risk probabilities and further re-rank the features. Finally, the predicted risks are calibrated. Together with our Australian partner, we perform comprehensive study on data collected for the mental health cohort, and the experiments validate that our proposed framework outperforms risk assessment instruments by medical practitioners.

Prevalence of disability in Australian elderly: impact of trends in obesity and diabetes

OBJECTIVE: We aimed to estimate the impact of past and future changes in obesity and diabetes prevalence in mid-life on disability prevalence for adult Australians. METHODS: We analysed data from the Australian Diabetes, Obesity and Lifestyle study (AusDiab) including participants aged 45-64years, disability-free at baseline (1999/2000) with disability information at follow-up (2011/12) (n=2107). We used coefficients from multinomial logistic regression to predict 10-year probabilities of disability and death from baseline predictors (age, sex, obesity, smoking, diabetes and hypertension). We estimated the prevalence of disability attributable to past (1980) and expected future (2025) changes in obesity and diabetes prevalence using the life table approach. RESULTS: We estimated that the prevalence of disability for those aged between 55 and 74years would have been 1697 cases per 100,000 persons less in 2010 (10.3% less) if the rates of obesity and diabetes observed in 2000 had been as low as the levels observed in 1980. However, if instead the prevalence of obesity and diabetes had been as high as the levels expected in 2025, then the prevalence of disability would have been an additional 2173 per 100,000 persons (an additional 13.2%). CONCLUSIONS: We demonstrate, for the first time, a substantial potential impact of obesity and diabetes trends on disability amongst those aged 55-74years. In Australian adults by 2025 we estimate that around 26% of disability cases would have been avoidable if there had been no change in obesity and diabetes prevalence since 1980. A similar impact is likely around the world in developed countries.

Can camera traps monitor Komodo dragons a large ectothermic predator?

Camera trapping has greatly enhanced population monitoring of often cryptic and low abundance apex carnivores. Effectiveness of passive infrared camera trapping, and ultimately population monitoring, relies on temperature mediated differences between the animal and its ambient environment to ensure good camera detection. In ectothermic predators such as large varanid lizards, this criterion is presumed less certain. Here we evaluated the effectiveness of camera trapping to potentially monitor the population status of the Komodo dragon (Varanus komodoensis), an apex predator, using site occupancy approaches. We compared site-specific estimates of site occupancy and detection derived using camera traps and cage traps at 181 trapping locations established across six sites on four islands within Komodo National Park, Eastern Indonesia. Detection and site occupancy at each site were estimated using eight competing models that considered site-specific variation in occupancy (ψ)and varied detection probabilities (p) according to detection method, site and survey number using a single season site occupancy modelling approach. The most parsimonious model [ψ (site), p (site survey); ω = 0.74] suggested that site occupancy estimates differed among sites. Detection probability varied as an interaction between site and survey number. Our results indicate that overall camera traps produced similar estimates of detection and site occupancy to cage traps, irrespective of being paired, or unpaired, with cage traps. Whilst one site showed some evidence detection was affected by trapping method detection was too low to produce an accurate occupancy estimate. Overall, as camera trapping is logistically more feasible it may provide, with further validation, an alternative method for evaluating long-term site occupancy patterns in Komodo dragons, and potentially other large reptiles, aiding conservation of this species.

Life history and the ecology of stress: how do glucocorticoid hormones influence life-history variation in animals?

Glucocorticoids hormones (GCs) are intuitively important for mediation of age-dependent vertebrate life-history transitions through their effects on ontogeny alongside underpinning variation in life-history traits and trade-offs in vertebrates. These concepts largely derive from the ability of GCs to alter energy allocation, physiology and behaviour that influences key life-history traits involving age-specific life-history transitions, reproduction and survival. Studies across vertebrates have shown that the neuroendocrine stress axis plays a role in the developmental processes that lead up to age-specific early life-history transitions. While environmental sensitivity of the stress axis allows for it to modulate the timing of these transitions within species, little is known as to how variation in stress axis function has been adapted to produce interspecific variation in the timing of life-history transitions. Our assessment of the literature confirms that of previous reviews that there is only equivocal evidence for correlative or direct functional relationships between GCs and variation in reproduction and survival. We conclude that the relationships between GCs and life-history traits are complex and general patterns cannot be easily discerned with current research approaches and experimental designs. We identify several future research directions including: (i) integration of proximate and ultimate measures, including longitudinal studies that measure effects of GCs on more than one life-history trait or in multiple environmental contexts, to test explicit hypotheses about how GCs and life-history variation are related and (ii) the measurement of additional factors that modulate the effects of GCs on life-history traits (e.g. GC receptors and binding protein levels) to better infer neurendocrine stress axis actions. Conceptual models of HPA/I axis actions, such as allostatic load and reactive scope, to some extent explicitly predict the role of GCs in a life-history context, but are descriptive in nature. We propose that GC effects on life-history transitions, survival probabilities and fecundity can be modelled in existing quantitative demographic frameworks to improve our understanding of how GC variation influences life-history evolution and GC-mediated effects on population dynamics Lay Summary Functional Ecology

Mammographic screening: measurement of the cost in a population based programme in Victoria, Australia

STUDY OBJECTIVES: To estimate the cost per woman participating in a mammographic screening programme, and to describe methods for measuring costs. DESIGN: Expenditure, resource usage, and throughput were monitored over a 12 month period. Unit costs for each phase of the screening process were estimated and linked with the probabilities of each screening outcome to obtain the cost per woman screened and the cost per breast cancer detected. SETTING: A pilot, population based Australian programme offering free two-view mammographic screening. PARTICIPANTS: A total of 5986 women aged 50-69 years who lived in the target area, were listed on the electoral roll, had no previous breast cancer, and attended the programme. RESULTS: Unit costs for recruitment, screening, and recall mammography were $17.54, $60.04, and $175.54, respectively. The costs of clinical assessment for women with subsequent clear, benign, malignant (palpable), and malignant (impalpable) diagnoses were $173.71, $527.29, $436.62, and $567.22, respectively. The cost per woman screened was $117.70, and the cost per breast cancer detected was $11,550. CONCLUSIONS: The cost per woman screened is a key variable in assessment of the cost effectiveness of mammographic screening, and is likely to vary between health care settings. Its measurement is justified if decisions about health care services are to be based on cost effectiveness criteria.

Detecting stepping stones by abnormal causality probability

Locating the real source of the Internet attacks has long been an important but difficult problem to be addressed. In the real world, attackers can easily hide their identities and evade punishment by relaying their attacks through a series of compromised systems or devices called stepping stones. Currently, researchers mainly use similar features from the network traffic, such as packet timestamps and frequencies, to detect stepping stones. However, these features can be easily destroyed by attackers using evasive techniques. In addition, it is also difficult to implement an appropriate threshold of similarity that can help justify the stepping stones. In order to counter these problems, in this paper, we introduce the consistent causality probability to detect the stepping stones. We formulate the ranges of abnormal causality probabilities according to the different network conditions, and on the basis of it, we further implement to self-adaptive methods to capture stepping stones. To evaluate our proposed detection methods, we adopt theoretic analysis and empirical studies, which demonstrate accuracy of the abnormal causality probability. Moreover, we compare our proposed methods with previous works. The result shows that our methods in this paper significantly outperform previous works in the accuracy of detection malicious stepping stones, even when evasive techniques are adopted by attackers.