Complex Statistical Models Research Articles

Challenges and Opportunities for the Agricultural Producers in Sinana District in Reflection of COVID-19 Pandemic

Introduction. Analyzing the effect of COVID-19 is an important issue in agricultural sectors. However, such analysis requires a complex hierarchical statistical model. Rapid spread of the COVID-19 pandemic has disrupted the world’s production and productivity in many sectors. Among those sectors, the agricultural sector is highly affected. The Bale zone in the larger extent and Sinana district, in particular, is one of the potential agricultural areas in the Oromia regional state, Ethiopia where agriculture is the major sector in supporting the livelihood of thousands of subsistence farmers in the area as well as the country at large. Research Methodology. This study involved primary data collected from the farmers in the Sinana district during the period 2020–2021. A total of 991 farmers were selected from the entire 22 kebeles in the district. The data were analyzed using multilevel binary logistic random intercept regression models with maximum-likelihood parameter estimation. Results. Of the 991 farmers, 549 (55.4%) responded that COVID-19 has brought only challenges in their agricultural production and 311 (31.4%) responded both challenges and opportunities. About 632 (63.8%) of the farmers said that there was wastage of products such as milk, dairy, fruits, and vegetables. Three hundred twenty-eight (33.1%) of the participants obtained modernization in their agricultural production system like use of tractors and irrigation systems. According to the model results, farmer’s sex, age, educational level, family size, farmland size, types of effect, aggravation in food insecurity, input delay, lack of workers, slowdown of service, falling in income, modernization in the system of production, wastage of product, and types of wasted products were identified as significant factors. About 8% of the total variability in the effect of COVID-19 is due to differences across kebeles (ICC = 0.08, P value ≤0.05), and the remaining is due to individual differences. Conclusion. This study further demonstrated the potential of a hierarchical model for the study of COVID-19 effect variation within and between the kebeles. The majority, about 92% variation in the effect, is due to the disparity of individuals (farmers). The farmers with large family sizes and high capacity to produce and who were females were negatively related to the effect of COVID-19 in agricultural production.

Meta-analytic approaches and effect sizes to account for 'nuisance heterogeneity' in comparative physiology.

Meta-analysis is a powerful tool used to generate quantitatively informed answers to pressing global challenges. By distilling data from broad sets of research designs and study systems into standardised effect sizes, meta-analyses provide physiologists with opportunities to estimate overall effect sizes and understand the drivers of effect variability. Despite this ambition, research designs in the field of comparative physiology can appear, at the outset, as being vastly different to each other because of 'nuisance heterogeneity' (e.g. different temperatures or treatment dosages used across studies). Methodological differences across studies have led many to believe that meta-analysis is an exercise in comparing 'apples with oranges'. Here, we dispel this myth by showing how standardised effect sizes can be used in conjunction with multilevel meta-regression models to both account for the factors driving differences across studies and make them more comparable. We assess the prevalence of nuisance heterogeneity in the comparative physiology literature - showing it is common and often not accounted for in analyses. We then formalise effect size measures (e.g. the temperature coefficient, Q10) that provide comparative physiologists with a means to remove nuisance heterogeneity without the need to resort to more complex statistical models that may be harder to interpret. We also describe more general approaches that can be applied to a variety of different contexts to derive new effect sizes and sampling variances, opening up new possibilities for quantitative synthesis. By using effect sizes that account for components of effect heterogeneity, in combination with existing meta-analytic models, comparative physiologists can explore exciting new questions while making results from large-scale data sets more accessible, comparable and widely interpretable.

Assessing the depression risk in the U.S. adults using nomogram

BackgroundDepression has received a lot of attention as a common and serious illness. However, people are rarely aware of their current depression risk probabilities. We aimed to develop and validate a predictive model applicable to the risk of depression in US adults.MethodsThis study was conducted using the database of the National Health and Nutrition Examination Survey (NHANES, 2017–2012). In particular, NHANES (2007–2010) was used as the training cohort (n = 6015) for prediction model construction and NHANES (2011–2012) was used as the validation cohort (n = 2812) to test the model. Depression was assessed (defined as a binary variable) by the Patient Health Questionnaire (PHQ-9). Socio-demographic characteristics, sleep time, illicit drug use and anxious days were assessed using a self-report questionnaire. Logistic regression analysis was used to evaluate independent risk factors for depression. The nomogram has the advantage of being able to visualize complex statistical prediction models as risk estimates of individualized disease probabilities. Then, we developed two depression risk nomograms based on the results of logistic regression. Finally, several validation methods were used to evaluate the prediction performance of nomograms.ResultsThe predictors of model 1 included gender, age, income, education, marital status, sleep time and illicit drug use, and model 2, furthermore, included anxious days. Both model 1 and model 2 showed good discrimination ability, with a bootstrap-corrected C index of 0.71 (95% CI, 0.69–0.73) and 0.85 (95% CI, 0.83–0.86), and an externally validated C index of 0.71 (95% CI, 0.68–0.74) and 0.83 (95% CI, 0.81–0.86), respectively, and had well-fitted calibration curves. The area under the receiver operating characteristic curve (AUC) values of the models with 1000 different weighted random sampling and depression scores of 10–17 threshold range were higher than 0.7 and 0.8, respectively. Calculated net reclassification improvement (NRI) and integrated discrimination improvement (IDI) showed the discrimination or accuracy of the prediction models. Decision curve analysis (DCA) demonstrated that the depression models were practically useful. The network calculators work for participants to make personalized predictions.ConclusionsThis study presents two prediction models of depression, which can effectively and accurately predict the probability of depression as well as helping the U.S. civilian non-institutionalized population to make optimal treatment decisions.

Power Modeling for Effective Datacenter Planning and Compute Management

Over the past decade, there has been a global growth in datacenter capacity, power consumption and the associated costs. Accurate mapping of datacenter resource usage (CPU, RAM, etc.) and hardware configurations (servers, accelerators, etc.) to its power consumption is necessary for efficient long-term infrastructure planning and real-time compute load management. This paper presents two types of statistical power models that relate CPU usage of Google’s Power Distribution Units (PDUs, commonly referred to as power domains) to their power consumption. The models are deployed in production and are used for cost- and carbon-aware load management, power provisioning and infrastructure rightsizing. They are simple, interpretable and exhibit uniformly high prediction accuracy in modeling power domains with large diversity of hardware configurations and workload types across Google fleet. A multi-year validation of the deployed models demonstrate that they can predict power with less than 5% Mean Absolute Percent Error (MAPE) for more than 95% diverse PDUs across Google fleet. This performance matches the best reported accuracies coming from studies that focus on specific workload types, hardware platforms and, typically, more complex statistical models.

Conceptual challenges for interpretable machine learning

As machine learning has gradually entered into ever more sectors of public and private life, there has been a growing demand for algorithmic explainability. How can we make the predictions of complex statistical models more intelligible to end users? A subdiscipline of computer science known as interpretable machine learning (IML) has emerged to address this urgent question. Numerous influential methods have been proposed, from local linear approximations to rule lists and counterfactuals. In this article, I highlight three conceptual challenges that are largely overlooked by authors in this area. I argue that the vast majority of IML algorithms are plagued by (1) ambiguity with respect to their true target; (2) a disregard for error rates and severe testing; and (3) an emphasis on product over process. Each point is developed at length, drawing on relevant debates in epistemology and philosophy of science. Examples and counterexamples from IML are considered, demonstrating how failure to acknowledge these problems can result in counterintuitive and potentially misleading explanations. Without greater care for the conceptual foundations of IML, future work in this area is doomed to repeat the same mistakes.

Building a Framework for Mode Effect Estimation in United States Presidential Election Polls

Abstract As public confidence in polling has been waning in the wake of recent elections (Narea, N. 2016. After 2016, Can we Ever Trust the Polls Again? New Republic), many researchers have been seeking to diagnose the shortcomings in these data (Gelman, A., and J. Azari. 2017. “19 Things We Learned from the 2016 Election.” Statistics and Public Policy 4 (1): 1–10; Kennedy, C., M. Blumenthal, S. Clement, J. D. Clinton, C. Durand, C. Franklin, K. McGeeney, L. Miringoff, K. Olson, D. Rivers, L. Saad, G. E. Witt, and C. Wlezien. 2018. “An Evaluation of the 2016 Election Polls in the United States.” Public Opinion Quarterly 82 (1): 1–33; Mercer, A., C. Deane, and K. McGeeney. 2016. Why 2016 Election Polls Missed Their Mark. Pew Research Center. Also available at https://www.pewresearch.org/fact-tank/2016/11/09/why-2016-election-polls-missed-their-mark/). One conjecture stems from observed differences between polling results based on the methodological choice between live and non-live modes of survey administration (Enten, H. 2015. The Future of Polling May Depend on Donald Trumps Fate. FiveThirtyEight. Also available at https://fivethirtyeight.com/features/the-future-of-polling-may-depend-on-donald-trumps-fate/). While it has become commonplace to discuss “mode effect” on surveys, it reemerged in the political zeitgeist as the “Shy Trump” supporter hypothesis leading up to the 2016 U.S. Presidential Election (Edsall, T. B. 2016. How Many People Support Trump but Dont want to Admit it. New York Times). Motivated by the conflicting evidence for (Enns, P. K., J. Lagodny, and J. P. Schuldt. 2017. “Understanding the 2016 US Presidential Polls: The Importance of Hidden Trump Supporters.” Statistics, Politics, and Policy 8 (1): 41–63) and against (Coppock, A. 2017. “Did Shy Trump Supporters Bias the 2016 Polls? Evidence from a Nationally-Representative List Experiment.” Statistics, Politics, and Policy 8 (1): 29–40) this hypothesis, we built a complex statistical model that pools together results across multiple pollsters and throughout the election cycle while accounting for the nuances of these data. Specifically, we explored election data for the presence of mode effect using time series with a general additive mixed model (GAMM). We estimated mode effect at state and national levels to perform statistical mode adjustments, which we then compared to observed election results. In this paper, we utilized polling results from the United States Presidential Elections in 2016 (4208 polls) and 2020 (4133 polls). Using these data, we identified spatial trends and areas where mode effect was statistically significant at a 0.05 level. In summary, we make three contributions to the literature on mode effect adjustment in the poll aggregation setting. First, we present a straightforward and flexible statistical approach to estimating mode effect using time series data. In doing so, we help to bridge the gap between theory-focused statistical work and the social sciences. Second, we apply this method to two recent presidential elections, providing insight into the significance of mode effect. Third, we provide evidence for spatial mode effect trends suggesting regional voting behaviors that future scholars can explore.

The burden of the most common rheumatic disease in Colombia

BackgroundEstimating the burden of rheumatic diseases (RDs) requires proper evaluation of its lethal and nonlethal consequences. In Colombia, it is possible to find local data and Global Burden of Disease (GBD) reports that collect information from varied contexts and apply complex statistical models, but no on-site estimations are available.MethodsThis was a descriptive study on the burden of RD based on occurrence and mortality data in the general population during 2015, including information and prevalence estimations from the Community Oriented Program for the Control of Rheumatic Diseases (COPCORD) study. Disability-adjusted life years (DALYs) were estimated by combining measures of years of life lost (YLL) and years lived with disability (YLDs). For disability weight estimations among cases, different COPCORD responses were mapped using flowcharts to show the severity distribution according to GBD. All model parameters and results were validated through an expert consensus panel.ResultsLow back pain (LBP) was the RD with the greatest burden of disease, costing 606.05 (95% CI 502.76–716.58) DALYs per 100,000 inhabitants, followed by osteoarthritis (292.11; 95% CI 205.76–386.85) and rheumatoid arthritis (192.46, 95% CI 109.7–239.69).ConclusionsThe burden of RD is as high in Colombia as in other countries of the region. The results offer an interesting tool for optimizing healthcare system design as well as for planning the distribution of human and economic resources to achieve early diagnosis and adequate care of these diseases.

Machine learning for microfluidic design and control.

Microfluidics has developed into a mature field with applications across science and engineering, having particular commercial success in molecular diagnostics, next-generation sequencing, and bench-top analysis. Despite its ubiquity, the complexity of designing and controlling custom microfluidic devices present major barriers to adoption, requiring intuitive knowledge gained from years of experience. If these barriers were overcome, microfluidics could miniaturize biological and chemical research for non-experts through fully-automated platform development and operation. The intuition of microfluidic experts can be captured through machine learning, where complex statistical models are trained for pattern recognition and subsequently used for event prediction. Integration of machine learning with microfluidics could significantly expand its adoption and impact. Here, we present the current state of machine learning for the design and control of microfluidic devices, its possible applications, and current limitations.

Moving Object Detection over Wireless Visual Sensor Networks using Spectral Dual Mode Background Subtraction

Wireless Visual Sensor Networks (WVSN) play an essential role in tracking moving objects. WVSN's key drawbacks are storage, power, and bandwidth. Background subtraction is used in the early stages of target tracking to extract moving targets from video images. Many standard methods of subtracting backgrounds are no longer suitable for embedded devices because they use complex statistical models to manage small changes in lighting. This paper introduces a system based on the Partial Discrete Cosine Transform (PDCT), reducing the vast dimensions of processed data while retaining most of the important information, thereby reducing processing and transmission energy. It also uses a dual-mode single Gaussian model (SGM) for accurate detection of moving objects. The proposed system's performance is to be assessed using the standard CDnet 2014 benchmark dataset in terms of detection accuracy and time complexity. Furthermore, the suggested method is compared to previous WVSN background subtraction methods. Simulation results show that the proposed method consistently has 15% better accuracy and is up to 3 times faster than the state-of-the-art object detection methods for WVSN. Finally, we showed the practicality of the suggested method by simulating it in a sensor network environment using the Contiki OS Cooja Simulator and implementing it in a real testbed using Cortex M3 open nodes of IOT-LAB.

Challenges for Estimating the Global Prevalence of Micronutrient Deficiencies and Related Disease Burden: A Case Study of the Global Burden of Disease Study.

ABSTRACTInformation on the prevalence of micronutrient deficiencies is needed to determine related disease burden; underpin evidence-based advocacy; and design, deliver, and monitor safe, effective interventions. Assessing the global prevalence of deficiency requires a valid micronutrient status biomarker with an appropriate cutoff to define deficiency and relevant data from representative surveys across multiple locations and years. The Global Burden of Disease Study includes prevalence estimates for iodine, iron, zinc, and vitamin A deficiencies, for which recommended biomarkers and appropriate deficiency cutoffs exist. Because representative survey data are lacking, only retinol concentration is used to model vitamin A deficiency, and proxy indicators are used for the other micronutrients (goiter for iodine, hemoglobin for iron, and dietary food adequacy for zinc). Because of data limitations, complex statistical modeling is required to produce current estimates, relying on assumptions and proxies that likely understate the extent of micronutrient deficiencies and the consequent global health burden.

Rainfall prediction: A comparative analysis of modern machine learning algorithms for time-series forecasting

Rainfall forecasting has gained utmost research relevance in recent times due to its complexities and persistent applications such as flood forecasting and monitoring of pollutant concentration levels, among others. Existing models use complex statistical models that are often too costly, both computationally and budgetary, or are not applied to downstream applications. Therefore, approaches that use Machine Learning algorithms in conjunction with time-series data are being explored as an alternative to overcome these drawbacks. To this end, this study presents a comparative analysis using simplified rainfall estimation models based on conventional Machine Learning algorithms and Deep Learning architectures that are efficient for these downstream applications. Models based on LSTM, Stacked-LSTM, Bidirectional-LSTM Networks, XGBoost, and an ensemble of Gradient Boosting Regressor, Linear Support Vector Regression, and an Extra-trees Regressor were compared in the task of forecasting hourly rainfall volumes using time-series data. Climate data from 2000 to 2020 from five major cities in the United Kingdom were used. The evaluation metrics of Loss, Root Mean Squared Error, Mean Absolute Error, and Root Mean Squared Logarithmic Error were used to evaluate the models’ performance. Results show that a Bidirectional-LSTM Network can be used as a rainfall forecast model with comparable performance to Stacked-LSTM Networks. Among all the models tested, the Stacked-LSTM Network with two hidden layers and the Bidirectional-LSTM Network performed best. This suggests that models based on LSTM-Networks with fewer hidden layers perform better for this approach; denoting its ability to be applied as an approach for budget-wise rainfall forecast applications.

An Internet of Things Platform Based on Microservices and Cloud Paradigms for Livestock.

With the growing adoption of the Internet of Things (IoT) technology in the agricultural sector, smart devices are becoming more prevalent. The availability of new, timely, and precise data offers a great opportunity to develop advanced analytical models. Therefore, the platform used to deliver new developments to the final user is a key enabler for adopting IoT technology. This work presents a generic design of a software platform based on the cloud and implemented using microservices to facilitate the use of predictive or prescriptive analytics under different IoT scenarios. Several technologies are combined to comply with the essential features—scalability, portability, interoperability, and usability—that the platform must consider to assist decision-making in agricultural 4.0 contexts. The platform is prepared to integrate new sensor devices, perform data operations, integrate several data sources, transfer complex statistical model developments seamlessly, and provide a user-friendly graphical interface. The proposed software architecture is implemented with open-source technologies and validated in a smart farming scenario. The growth of a batch of pigs at the fattening stage is estimated from the data provided by a level sensor installed in the silo that stores the feed from which the animals are fed. With this application, we demonstrate how farmers can monitor the weight distribution and receive alarms when high deviations happen.

Probabilistic Flood Hazard Maps from Monte Carlo Derived Peak Flow Values—An Application to Flood Risk Management in Zamora City (Spain)

All flood hazard and risk assessment suffer from a certain degree of uncertainty due to multiple factors, such as flood frequency analysis, hydrodynamic model calibration, or flood damage (magnitude–damage functions) models. The uncertainty linked to the flood frequency analysis is one of the most important factors (previous and present estimation point to 40%). Flood frequency analysis uncertainty has been approached from different points of view, such as the application of complex statistical models, the regionalization processes of peak flows, or the inclusion of non-systematic data. Here, we present an achievable approach to defining the uncertainty linked to flood frequency analysis by using the Monte Carlo method. Using the city of Zamora as the study site, the uncertainty is delimited by confidence intervals of a peak flow quantile of a 500-year return period. Probabilistic maps are derived from hydrodynamic results, and further analysis include flood hazard maps for human loss of stability and vehicle damage. Although the effect of this uncertainty is conditioned by the shape of the terrain, the results obtained may allow managers to achieve more consistent land-use planning. All those Zamora city results point out the probable underestimation of flood hazard (the higher hazard areas increase around 20%) and risk when the uncertainty analysis is not considered, thus limiting the efficiency of flood risk management tasks.

IMI2-PainCare-BioPain-RCT3: a randomized, double-blind, placebo-controlled, crossover, multi-center trial in healthy subjects to investigate the effects of lacosamide, pregabalin, and tapentadol on biomarkers of pain processing observed by electroencephalography (EEG)

BackgroundIMI2-PainCare-BioPain-RCT3 is one of four similarly designed clinical studies aiming at profiling a set of functional biomarkers of drug effects on the nociceptive system that could serve to accelerate the future development of analgesics, by providing a quantitative understanding between drug exposure and effects of the drug on nociceptive signal processing in human volunteers. IMI2-PainCare-BioPain-RCT3 will focus on biomarkers derived from non-invasive electroencephalographic (EEG) measures of brain activity.MethodsThis is a multisite single-dose, double-blind, randomized, placebo-controlled, 4-period, 4-way crossover, pharmacodynamic (PD) and pharmacokinetic (PK) study in healthy subjects. Biomarkers derived from scalp EEG measurements (laser-evoked brain potentials [LEPs], pinprick-evoked brain potentials [PEPs], resting EEG) will be obtained before and three times after administration of three medications known to act on the nociceptive system (lacosamide, pregabalin, tapentadol) and placebo, given as a single oral dose in separate study periods. Medication effects will be assessed concurrently in a non-sensitized normal condition and a clinically relevant hyperalgesic condition (high-frequency electrical stimulation of the skin). Patient-reported outcomes will also be collected. A sequentially rejective multiple testing approach will be used with overall alpha error of the primary analysis split between LEP and PEP under tapentadol. Remaining treatment arm effects on LEP or PEP or effects on EEG are key secondary confirmatory analyses. Complex statistical analyses and PK-PD modeling are exploratory.DiscussionLEPs and PEPs are brain responses related to the selective activation of thermonociceptors and mechanonociceptors. Their amplitudes are dependent on the responsiveness of these nociceptors and the state of the pathways relaying nociceptive input at the level of the spinal cord and brain. The magnitude of resting EEG oscillations is sensitive to changes in brain network function, and some modulations of oscillation magnitude can relate to perceived pain intensity, variations in vigilance, and attentional states. These oscillations can also be affected by analgesic drugs acting on the central nervous system. For these reasons, IMI2-PainCare-BioPain-RCT3 hypothesizes that EEG-derived measures can serve as biomarkers of target engagement of analgesic drugs for future Phase 1 clinical trials. Phase 2 and 3 clinical trials could also benefit from these tools for patient stratification.Trial registrationThis trial was registered 25/06/2019 in EudraCT (2019%2D%2D001204-37).

Predicting exoplanet mass from radius and incident flux: a Bayesian mixture model

ABSTRACT The relationship between mass and radius (M–R relation) is the key for inferring the planetary compositions and thus valuable for the studies of formation and migration models. However, the M–R relation alone is not enough for planetary characterization due to the dependence of it on other confounding variables. This paper provides a non-trivial extension of the M–R relation by including the incident flux as an additional variable. By using Bayesian hierarchical modelling (BHM) that leverages the flexibility of finite mixture models, a probabilistic mass–radius–flux relationship (M–R–F relation) is obtained based on a sample of 319 exoplanets. We find that the flux has non-negligible impact on the M–R relation, while such impact is strongest for hot Jupiters. On the population level, the planets with higher level of flux tend to be denser, and high flux could trigger significant mass loss for plants with radii larger than 13R⊕. As a result, failing to account for the flux in mass prediction would cause systematic over- or underestimation. With the recent advent of computing power, although a lot of complex statistical models can be fitted using Monte Carlo methods, it has largely remained illusive how to validate these complex models when the data are observed with large measurement errors. We present two novel methods to examine model assumptions, which can be used not only for the models we present in this paper but can also be adapted for other statistical models.

Deep neural networks for choice analysis: A statistical learning theory perspective

Although researchers increasingly use deep neural networks (DNN) to analyze individual choices, overfitting and interpretability issues remain obstacles in theory and practice. This study presents a statistical learning theoretical framework to examine the tradeoff between estimation and approximation errors, and between the quality of prediction and of interpretation. It provides an upper bound on the estimation error of the prediction quality in DNN, measured by zero-one and log losses, shedding light on why DNN models do not overfit. It proposes a metric for interpretation quality by formulating a function approximation loss that measures the difference between true and estimated choice probability functions. It argues that the binary logit (BNL) and multinomial logit (MNL) models are the specific cases of DNNs, since the latter always has smaller approximation errors. We explore the relative performance of DNN and classical choice models through three simulation scenarios comparing DNN, BNL, and binary mixed logit models (BXL), as well as one experiment comparing DNN to BNL, BXL, MNL, and mixed logit (MXL) in analyzing the choice of trip purposes based on the National Household Travel Survey 2017. The results indicate that DNN can be used for choice analysis beyond the current practice of demand forecasting because it has the inherent utility interpretation and the power of automatically learning utility specification. Our results suggest DNN outperforms BNL, BXL, MNL, and MXL models in both prediction and interpretation when the sample size is large (≥O(104)), the input dimension is high, or the true data generating process is complex, while performing worse when the opposite is true. DNN outperforms BNL and BXL in zero-one, log, and approximation losses for most of the experiments, and the larger sample size leads to greater incremental value of using DNN over classical discrete choice models. Overall, this study introduces the statistical learning theory as a new foundation for high-dimensional data, complex statistical models, and non-asymptotic data regimes in choice analysis, and the experiments show the effective prediction and interpretation of DNN for its applications to policy and behavioral analysis.

Asymptotics of maximum likelihood estimators based on Markov chain Monte Carlo methods

In complex statistical models, in which exact computation of the likelihood is intractable, Monte Carlo methods can be applied to approximate maximum likelihood estimates. In this paper we consider approximation obtained via Markov chain Monte Carlo. We prove consistency and asymptotic normality of the resulting estimator, when both sample sizes (the initial and Monte Carlo one) tend to infinity. Our results can be applied to models with intractable normalizing constants and missing data models. We also investigate properties of estimators in numerical experiments.

Bayesian regression models for ecological count data in PyMC3

Ecological inferences need structurally flexible statistical models to accommodate complex ecological phenomena. PyMC3 is a Probabilistic Programming Language (PPL) and allows for custom statistical distributions to build complex statistical models. This study used PyMC3 to implement Bayesian generalized Poisson (GP), zero-inflated GP, and hurdle GP regression models for over- and under-dispersed counts. The Bayesian GP regression models were fitted to simulated counts and real-world counts of over- and under-dispersion, respectively. Coefficient estimates of the Bayesian regression models were consistent with the known values used in the simulations and those of published work or models. Simulations demonstrated that Bayesian GP regression models with the NUTS sampler worked correctly for under-dispersed counts if the number of non-zero frequency classes was five or more. PyMC3 is not only flexible for building complex statistical models using custom likelihood functions, but also syntactically concise. The programming flexibility of PyMC3 can provide ecologists and environmental scientists with flexible, robust Bayesian computational platforms.

Assessment of West Nile Virus Lineage 2 Dynamics in Greece and Future Implications.

The emergence of West Nile Virus lineage 2 (WNV-2) has contributed to multiple major human outbreaks in Greece since 2010. Studies to date investigating biological and environmental factors that contribute to West Nile Virus (WNV) transmission have resulted in complex statistical models. We sought to examine open publicly available data to ascertain if a predictive risk assessment could be employed for WNV-2 in Greece. Based on accessible data, factors such as precipitation, temperature, and range of avian host species did not yield conclusive outcomes. However, by measuring the average rate of temperature change leading up to peak caseloads, we found a predictive characteristic to the timing of outbreaks. Detailed evolutionary studies revealed possible multiple introductions of WNV-2 in Europe, and that Greece acts through a source-sink inversion model, thereby allowing continued reseeding of WNV transmission each year by overwintering the Culex pipiens mosquito vector. Greece has proven an excellent model in WNV surveillance and has demonstrated the importance of rapid reporting for proper preparedness and response to vector-borne diseases.

Artificial Intelligence models for prediction of the tide level in Venice

The city of Venice is an extraordinary architectural, artistic and cultural heritage. Unfortunately, its conservation is increasingly threatened by particularly significant high tides. Predicting the tide level in Venice, especially the high waters, is an essential task for the protection of the city and the lagoon. Complex statistical or hydrodynamic models, which require a large amount of input data, are currently used for this purpose. An effective alternative can be provided by models based on Artificial Intelligence algorithms. In this study, several different forecasting models were developed and each model was built in three variants, varying the implemented machine learning algorithm: M5P Regression Tree, Random Forest and Multilayer Perceptron. Until now, regression tree models had never been used to forecast tide levels. All the proposed models proved to be able to forecast the tide level in Venice with good accuracy. The M5P algorithm provided the best performance in most cases. All the models based on M5P were characterized by a coefficient of determination between 0.924 and 0.996, while the Relative Absolute Error was between 5.98 and 26.84%. In addition, good predictions were achieved by neglecting meteorological factors, even in the case of exceptionally high waters. Finally, satisfactory outcomes were also obtained with a forecast horizon of several hours, while a further specific comparison showed that the models based on the considered Machine Learning algorithms are able to outperform the AutoRegressive Integrated Moving Average models with exogenous input variables in forecasting high water.