Specifying, assessing, and selecting among candidate statistical models is fundamental to ecological research. Commonly used approaches to model selection are based on predictive scores and include information criteria such as Akaike's information criterion, and cross validation. Based on data splitting, cross validation is particularly versatile because it can be used even when it is not possible to derive a likelihood (e.g., many forms of machine learning) or count parameters precisely (e.g., mixed‐effects models). However, much of the literature on cross validation is technical and spread across statistical journals, making it difficult for ecological analysts to assess and choose among the wide range of options. Here we provide a comprehensive, accessible review that explains important—but often overlooked—technical aspects of cross validation for model selection, such as: bias correction, estimation uncertainty, choice of scores, and selection rules to mitigate overfitting. We synthesize the relevant statistical advances to make recommendations for the choice of cross‐validation technique and we present two ecological case studies to illustrate their application. In most instances, we recommend using exact or approximate leave‐one‐out cross validation to minimize bias, or otherwise k‐fold with bias correction if k < 10. To mitigate overfitting when using cross validation, we recommend calibrated selection via our recently introduced modified one‐standard‐error rule. We advocate for the use of predictive scores in model selection across a range of typical modeling goals, such as exploration, hypothesis testing, and prediction, provided that models are specified in accordance with the stated goal. We also emphasize, as others have done, that inference on parameter estimates is biased if preceded by model selection and instead requires a carefully specified single model or further technical adjustments.
Southeast Asia is highly biodiverse and currently experiences among the highest rates of tropical deforestation globally, but impacts on biodiversity are not well synthesized. We use Bayesian multi-level modeling to meta-analyse 831 pairwise comparisons of biodiversity in sites subject to land-use driven deforestation (for example, plantations or logged forest) versus undisturbed sites (control sites). After controlling for hierarchical dependencies, we show that biodiversity is a fifth lower in sites with these land-use driven deforestation (95% credible interval = 16%–28%, mean = 22%). This reduction was greater when forest losses were of high-intensity (34% reduction in biodiversity) compared to low-intensity (18% reduction), and effects were consistent across biogeographic regions and taxa. Oil-palm plantations led to the greatest reduction in biodiversity (39%, CI 27%–48%), and agroforests the least (24%, CI 10%–37%). We also find that biodiversity was reduced by 26% (CI 4%–42%) in secondary forest sites compared to undisturbed control sites, but biodiversity was the same in intermediate or mature-aged secondary forest compared to control sites (although species composition was potentially altered). Overall, our study provides a new line of evidence of the substantial detrimental impacts of land-use driven deforestation and particular types of land-use on the biodiversity of Southeast Asia.
Monitoring wildlife is crucial for making informed conservation and land-management decisions. Remotely triggered cameras are widely used for this purpose, but the resulting 'big data' are laborious to process. Although artificial intelligence (AI) offers a powerful solution to this bottleneck, it has been challenging for ecologists and practitioners without substantial technical expertise to tailor current approaches to their specific use cases. Generic, online offerings also have issues of ongoing costs and data privacy. Here we present an open-source, scalable, modular, cross-platform workflow, deployed using Docker containers, which leverages deep learning for wildlife image classification. It can be run using simple command-line prompts or via a user-friendly graphical user interface (AddaxAI). It enables end-users to easily execute a full range of tasks—from animal detection and counting to species identification—on local or cloud GPU-accelerated machines. It also integrates with the widely used open-source camera-trapping software ‘Camelot’, writing AI-classification data directly to image metadata and to CSV files, ready for either expert verification or direct data analysis. The result is a user-friendly but powerful multi-platform application for wildlife-image classification and research pipelines. An example case study with Tasmanian wildlife demonstrates the utility of our classifier training and inference workflow.
19 April 2022 in Australasian Journal of Educational Technology
Student evaluation surveys provide educational institutions with important feedback regarding the student experience of teaching and courses; however, qualitative comments can contain offensive, insulting or threatening content. Large educational institutions generate thousands of comments per academic term; therefore, manual screening processes to find potentially harmful comments are not generally feasible. We developed a methodology for semi-automated screening of student comments that incorporates a machine learning decision support system and a detailed psychological assessment protocol. In a case study at a large public Australian university, our system identified 4,258 out of 62,049 (6.9%) comments as potentially harmful and requiring further review. Feedback from stakeholders demonstrates that this methodology is useful in reducing staff workload and could be broadly applied to different settings. Implications for practice or policy:
Rapid antigen testing (RAT) is a cost-effective and time-efficient method of identifying severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and therefore a critical part of infection control strategies. There is no published evidence evaluating the use of RAT during the Omicron wave of the COVID-19 pandemic for asymptomatic patients or its performance between waves. All patients presenting to an Emergency Department over a two-week period without COVID-19 symptoms were screened for SARS-CoV-2 using both the Abbott Panbio RAT as well as the gold standard reverse transcriptase real-time polymerase chain reaction (PCR). The Abbott Panbio RAT sensitivity was 13% (95% CI 0.028, 0.336) for asymptomatic patients. The use of this test in asymptomatic patients during the Omicron wave had a statistically significant reduction in sensitivity compared with two reports of the same test in previous waves (13% vs. 86%, p < 0.0001; 13% vs. 83%, p < 0.0001). As SARS-CoV-2 continues to mutate, the sensitivity of RATs are altered and needs to be continually re-evaluated for each variant of concern if they are to be used as part of an infection control strategy.
We analysed fossil mammal assemblages from over 350 Late Pleistocene and Holocene sites worldwide to test whether human activities, such as agriculture, domestication and intensified land use, restructured global patterns of mammal co-occurrence. Using presence–absence data, we contrasted a novel iterative ‘chase clustering’ method, which is compositionally driven, against a traditional spatially constrained Ward’s clustering approach. Both methods recovered continental-scale groupings in the Pleistocene, consistent with known biogeographic boundaries. Holocene land use and domestication reconfigured these historical patterns, creating novel assemblages independent of previous biogeographic constraints. Faunal turnover at the local scale varied substantially across regions, being especially pronounced in the Americas, whereas other areas showed relative stability. Even moderate expansion of domesticates altered how communities grouped, highlighting their disproportionate ecological influence. Our findings demonstrate that human-driven niche modification, beyond earlier megafaunal extinctions, profoundly reshaped mammal communities on a global scale. Recognizing these anthropogenic legacies provides essential context for anticipating how current and future human pressures might further transform biodiversity.