Small Training Research Articles

Updating Nonlinear Stochastic Dynamics of an Uncertain Nozzle Model Using Probabilistic Learning With Partial Observability and Incomplete Dataset

Abstract This article introduces a methodology for updating the nonlinear stochastic dynamics of a nozzle with uncertain computational model. The approach focuses on a high-dimensional nonlinear computational model constrained by a small target dataset. Challenges include the large number of degrees-of-freedom, geometric nonlinearities, material uncertainties, stochastic external loads, underobservability, and high computational costs. A detailed dynamic analysis of the nozzle is presented. An updated statistical surrogate model relating the observations of interest to the control parameters is constructed. Despite small training and target datasets and partial observability, the study successfully applies probabilistic learning on manifolds (PLoM) to address these challenges. PLoM captures geometric nonlinear effects and uncertainty propagation, improving conditional mean statistics compared to training data. The conditional confidence region demonstrates the ability of the methodology to accurately represent both observed and unobserved output variables, contributing to advancements in modeling complex systems.

A robust neural network for privacy-preserving heart rate estimation in remote healthcare systems

In this study, we propose a computationally-light and robust neural network for estimating heart rate in remote healthcare systems. We develop a model that can be trained on consumer-grade graphics processing units (GPUs), and can be deployed on edge devices for swift inference. We propose a hybrid model based on convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM) architectures for estimating heart rate from Electrocardiogram (ECG) and Photoplethysmography (PPG) signals. Considering the sensitive nature of the ECG signals, we ensure a formal privacy guarantee, differential privacy, for the model training. We perform a tight accounting of the overall privacy budget of our training algorithm using the Rényi Differential Privacy technique. We demonstrate that our model outperforms state-of-the-art networks on a benchmark dataset for both ECG and PPG signals despite having a much smaller number of trainable parameters and, consequently, much smaller training and inference times. Our CNN-BiLSTM architecture can also provide excellent heart rate estimation performance even under strict privacy constraints. We develop a prototype Arduino-based data collection system that is low-cost, efficient, and useful for providing access to modern healthcare services to people living in remote areas.

Tiny trains – new hope? Or the same issues in a smaller box?

The report card from Australia’s decade-long liquefied natural gas (LNG) construction boom makes for sobering reading. From 2009 to 2019, there were 15 new trains turned on, which lifted LNG production to a record 77 million tonnes in 2019. They are clearly big plants, with big production trains and with big capital expenditure to match. But the plants seemed to come with big problems. Not just in construction, but sometimes in production too. Now times have changed. Advances in technology, together with new market dynamics, have sparked new interest in small trains. A number of projects in the USA are now being proposed with multiple smaller units rather than the big trains that have dominated the last decade. For Australian owners and operators a new questions emerges. Do these ‘tiny trains’ represent a new step on the journey to a more efficient and effective commercialisation of region gas? Or will the new trains deliver the same issues, just in a smaller box? This paper provides: lessons learned from the Australian LNG boom between 2009 and 2019, an overview of the new ‘small train’ technologies including real world examples, a discussion of the risks and benefits of this approach, a case study from US owners/operators and discussion of where this approach might suit the needs of Australian LNG. The paper will appeal to oil and gas producers, major contractors and service providers.

Classification of Impact Echo Signals Using Explainable Deep Learning and Transfer Learning Approaches

Impact echo (IE) is one of the most frequently used nondestructive evaluation (NDE) techniques for detecting subsurface defects such as delamination, honeycombing, and voids in concrete structures. In the conventional analysis of IE data, the time-domain signal is transformed into the frequency domain and the frequency content is used to estimate the presence and nature of the defect. Machine learning (ML) has been recently applied to the IE signal classification problem. However, because of the scarcity of labeled IE datasets, most existing work relies on relatively small training and test datasets without addressing the generalizability and transferability of the developed models. In this paper, we compare two approaches for automatic classification of IE signals: clustering based on expert-crafted features and deep learning (DL) from automatically extracted features. Next, we use the knowledge gained from a DL model trained on concrete specimens with available ground truth to make predictions about defects in a different specimen with completely different construction and characteristics (transfer learning). Finally, we examine our DL model to gain insights into the model working (explainability) and highlight the attributions that are significant in classifying a particular IE signal. Our findings demonstrate the utility of ML and DL for IE signal classification, but also highlight the need for high-quality labeled datasets for advancing ML and DL in NDE data analysis.

Prediction of SO2 and NOx in sintering flue gas based on PSO-BP neural network model

ABSTRACT In the long process of iron and steel, the sintering process has the largest amount of flue gas emissions, many types of pollutants and high concentrations. The source control of SO2 and NOx in sintering flue gas through digital technology has become a new emission reduction technology. In this study, the BP neural network model (BP-NN) is optimized by using the particle swarm algorithm (PSO) to form the PSO-BPNN model, which effectively improves the characteristics of BP-NN with slow convergence speed and easily falls into local minima, and improves the learning ability and generalization. The test results show that the PSO-BP-NN algorithm not only has fast convergence speed and high prediction accuracy, but also has smaller training and inspection errors. In addition, this model combines process theory and feature engineering selection of parameters, which effectively improves the accuracy of the model and the interpretability of the results based on the linkage of process knowledge, and has certain analytical significance for the source management and post-treatment of sintered flue gas.

Sonified Distance in Sensory Substitution Does Not Always Improve Localization: Comparison With a 2-D and 3-D Handheld Device

Early visual to auditory substitution devices encode 2-D monocular images into sounds while more recent devices use distance information from 3-D sensors. This study assesses whether the addition of sound-encoded distance in recent systems helps to convey the “where” information. This is important to the design of new sensory substitution devices. We conducted experiments for object localization and navigation tasks with a handheld visual to audio substitution system. It comprises 2-D and 3-D modes. Both encode in real-time the position of objects in images captured by a camera. The 3-D mode encodes in addition the distance between the system and the object. Experiments have been conducted with 16 blindfolded sighted participants. For the localization, participants were quicker to understand the scene with the 3-D mode that encodes distances. On the other hand, with the 2-D only mode, they were able to compensate for the lack of distance encoding after a small training. For the navigation, participants were as good with the 2-D only mode than with the 3-D mode encoding distance.

Efficient contour-based annotation by iterative deep learning for organ segmentation from volumetric medical images

PurposeTraining deep neural networks usually require a large number of human-annotated data. For organ segmentation from volumetric medical images, human annotation is tedious and inefficient. To save human labour and to accelerate the training process, the strategy of annotation by iterative deep learning recently becomes popular in the research community. However, due to the lack of domain knowledge or efficient human-interaction tools, the current AID methods still suffer from long training time and high annotation burden.MethodsWe develop a contour-based annotation by iterative deep learning (AID) algorithm which uses boundary representation instead of voxel labels to incorporate high-level organ shape knowledge. We propose a contour segmentation network with a multi-scale feature extraction backbone to improve the boundary detection accuracy. We also developed a contour-based human-intervention method to facilitate easy adjustments of organ boundaries. By combining the contour-based segmentation network and the contour-adjustment intervention method, our algorithm achieves fast few-shot learning and efficient human proofreading.ResultsFor validation, two human operators independently annotated four abdominal organs in computed tomography (CT) images using our method and two compared methods, i.e. a traditional contour-interpolation method and a state-of-the-art (SOTA) convolutional network (CNN) method based on voxel label representation. Compared to these methods, our approach considerably saved annotation time and reduced inter-rater variabilities. Our contour detection network also outperforms the SOTA nnU-Net in producing anatomically plausible organ shape with only a small training set.ConclusionTaking advantage of the boundary shape prior and the contour representation, our method is more efficient, more accurate and less prone to inter-operator variability than the SOTA AID methods for organ segmentation from volumetric medical images. The good shape learning ability and flexible boundary adjustment function make it suitable for fast annotation of organ structures with regular shape.

The Study of Abstract Translation of Non-English Department Student at UPY

his study aims (1) to describe the problems of students when they have to translate an abstract from Indonesia to English, (2) to describe translation technique (TT) used by students when translating abstract, (3) to describe the impact of the application of TT used on the quality of abstract translation, (4) to describe the solution that the researcher takes to solve the problems in translating abstract. This study was conducted at University of PGRI Yogyakarta, 27th June 2021. It was qualitative study with total 10 respondents using FGD technique. They were 5 males and 5 females (four students of management, three students of history education, two students of civics education, and one student of mathematics education). Interviews and questionnaire in FGD were used to collect the data. The result showed that (1) the respondents have problems in translating abstract such as in choosing article, in understanding the English, in having less vocabulary, in knowing the meaning, in asking someone’s help to translate, in understanding grammar, in translating abstract without any help from their supervisor and in translating abstract using Google translate only. (2) The respondents used pure borrowing and established equivalent techniques. (3) The application of pure borrowing and established equivalent technique have impacted to the translation quality. Their abstract translation quality score C, D and E. (4) The researcher did the small training as the solution to improve their abstract quality.

VP-Detector: A 3D multi-scale dense convolutional neural network for macromolecule localization and classification in cryo-electron tomograms

Background and objective: Cryo-electron tomography (cryo-ET) with subtomogram averaging (STA) is indispensable when studying macromolecule structures and functions in their native environments. Due to the low signal-to-noise ratio, the missing wedge artifacts in tomographic reconstructions, and multiple macromolecules of varied shapes and sizes, macromolecule localization and classification remain challenging. To tackle this bottleneck problem for structural determination by STA, we design an accurate macromolecule localization and classification method named voxelwise particle detector (VP-Detector).Methods: VP-Detector is a two-stage particle detection method based on a 3D multiscale dense convolutional neural network (3D MSDNet). The proposed network uses 3D hybrid dilated convolution (3D HDC) to avoid the resolution loss caused by scaling operations. Meanwhile, it uses 3D dense connectivity to encourage the reuse of feature maps to reduce trainable parameters. In addition, the weighted focal loss is proposed to focus more attention on difficult samples and rare classes, which relieves the class imbalance caused by multiple particles of various sizes. The performance of VP-Detector is evaluated on both simulated and real-world tomograms, and it shows that VP-Detector outperforms state-of-the-art methods.Results: The experiments show that VP-Detector outperforms the state-of-the-art methods on particle localization with an F1-score of 0.951 and a precision of 0.978. In addition, VP-Detector can replace manual particle picking in experiment on the real-world tomograms. Furthermore, it performs well in classifying large-, medium-, and small-weight proteins with accuracies of 1, 0.95, and 0.82, respectively. Finally, ablation studies demonstrate the effectiveness of 3D HDC, 3D dense connectivity, weighted focal loss, and training on small training sets.Conclusions: VP-Detector can achieve high accuracy in particle detection with few trainable parameters and support training on small datasets. It can also relieve the class imbalance caused by multiple particles with various shapes and sizes.

On the Use of Morpho-Syntactic Description Tags in Neural Machine Translation with Small and Large Training Corpora

With the transition to neural architectures, machine translation achieves very good quality for several resource-rich languages. However, the results are still much worse for languages with complex morphology, especially if they are low-resource languages. This paper reports the results of a systematic analysis of adding morphological information into neural machine translation system training. Translation systems presented and compared in this research exploit morphological information from corpora in different formats. Some formats join semantic and grammatical information and others separate these two types of information. Semantic information is modeled using lemmas and grammatical information using Morpho-Syntactic Description (MSD) tags. Experiments were performed on corpora of different sizes for the English–Slovene language pair. The conclusions were drawn for a domain-specific translation system and for a translation system for the general domain. With MSD tags, we improved the performance by up to 1.40 and 1.68 BLEU points in the two translation directions. We found that systems with training corpora in different formats improve the performance differently depending on the translation direction and corpora size.

Application of logistic concept for the organization of small academic groups training in higher education institutions

Purpose. From the standpoint of educational logistics, we consider organization of training of small academic groups (SAG) in higher education institutions (HEIs). We consider separate components of the logistics system of HEIs and features of resource, information and service streams for maintenance of training of SAG. We substantiate the expediency of organizing student training in the SAG to meet regional staffing needs. Methodology. In our study, we use system-structural methods (including systems analysis, generalization, research-targeted generalization and systematization of economic, statistical, and scientific-methodological data) as well as the logistical approach. To substantiate the results, we conduct a sociological survey using the statistical data processing based on multicomponent correlation analysis. Findings. The institution of higher education is considered as a logistics system consisting of numerous interconnected links that require the use of information, resource and service flows adapted to educational conditions. For flow management, the use of logistics technologies is proposed, which are combined in the Learning and Development (L&D) model, which has four components. The authors provide a detailed review of these components with highlights on their functional aspects that aim to ensure an economically, organizationally and methodologically sound joint SAG training for students from different specializations. The data is provided testifying to expediency of carrying out training of experts in HEIs of that region where there is a need for these experts, even in spite of bigger cost of training of SAG students. Originality. For the first time, a logistical concept was used to consider the organization of SAG training in the HEIs. The application of the L&D model for logistics system flow management has been improved, as well as the contaminated form of training and diversification of subject content, which ensure the quality of the service flow during the joint training of students of different specialties. Practical value. The application of the logistics concept for the organization of SAG training in HEIs allows one, on the one hand, to optimally realize the state interests in regional staffing, and on the other to increase the stability of HEIs in a rapidly changing market economy.

Efficacy of the Fast Track Friendship Group Program for Peer-rejected Children: A Randomized-Controlled Trial

ABSTRACT Objective To evaluate the benefits of the Fast Track Friendship Group program implemented as a stand-alone school-based intervention on the social cognitions, social behavior, peer and teacher relationships of peer-rejected students. Method Over four successive years, 224 peer-rejected elementary students (57% White, 17% Black, 20% Latinx, 5% multiracial; 68% male; grades 1–4; Mage = 8.1 years old) were identified using peer sociometric nominations and randomly assigned to the intervention (n = 110) or a treatment-as-usual control group (n = 114). Four school districts serving economically-disadvantaged urban and rural communities participated. Intervention involved weekly small group social skills training with classmate partners, with sessions tailored to address individual child needs. Consultation meetings held at the start and mid-point of intervention were designed to help teachers and parents support the generalization of targeted skills. Results Multi-level linear models, with children nested within schools (controlling for demographics and baseline scores) documented improvements in social-cognitive skills (direct assessments of emotion recognition and competent social problem-solving), social behavior (teacher ratings of social skills and externalizing behavior), and interpersonal relationships (peer sociometric nominations of peer acceptance and friendships, teacher-rated student-teacher closeness). Significant effects were generally small (ds = .19–.36) but consistent across child sex, grade level, and behavioral characteristics. Conclusions The intervention proved feasible for high-fidelity implementation in school settings and produced significant improvements in the social adjustment of peer-rejected children, validating the approach as a school-based Tier 2 intervention.

Integrating Handcrafted Features with Deep Representations for Smartphone Authentication

Recent research demonstrates the potential of touch dynamics as a usable and privacy-preserving scheme for smartphone authentication. Most existing approaches rely on handcrafted features since deep models may be vulnerable to behavioral uncertainty due to the lack of consistent semantic information. Toward this end, we propose an approach to integrating handcrafted features into two phases of the deep learning process. On one hand, we present three fine-grained behavior representations by encoding semantic handcrafted features into the raw touch actions. On the other hand, we devise a deep Feature Regularization Net (FRN) architecture to combine the complementary information in both handcrafted and deep features. FRN involves handcrafted features as regularization to guide the learning process of deep features and selectively fuses these two feature types through a feature re-weighting mechanism. Experimental findings demonstrate that FRN outperforms the existing handcrafted or deep features even with smaller training and template sets. The framework also works for SOTA deep models and further boosts the accuracy. Results show that our approach is more reliable to alleviate behavioral variability and is competitively robust to statistical attacks compared with the most effective handcrafted features, suggesting a promising mechanism to improve the effectiveness and usability of behavioral authentication for multi-touch enabled mobile devices.

Cell segmentation for immunofluorescence multiplexed images using two-stage domain adaptation and weakly labeled data for pre-training

Cellular profiling with multiplexed immunofluorescence (MxIF) images can contribute to a more accurate patient stratification for immunotherapy. Accurate cell segmentation of the MxIF images is an essential step. We propose a deep learning pipeline to train a Mask R-CNN model (deep network) for cell segmentation using nuclear (DAPI) and membrane (Na+K+ATPase) stained images. We used two-stage domain adaptation by first using a weakly labeled dataset followed by fine-tuning with a manually annotated dataset. We validated our method against manual annotations on three different datasets. Our method yields comparable results to the multi-observer agreement on an ovarian cancer dataset and improves on state-of-the-art performance on a publicly available dataset of mouse pancreatic tissues. Our proposed method, using a weakly labeled dataset for pre-training, showed superior performance in all of our experiments. When using smaller training sample sizes for fine-tuning, the proposed method provided comparable performance to that obtained using much larger training sample sizes. Our results demonstrate that using two-stage domain adaptation with a weakly labeled dataset can effectively boost system performance, especially when using a small training sample size. We deployed the model as a plug-in to CellProfiler, a widely used software platform for cellular image analysis.

A Study on Prediction Skills and Reading Efficiency in College English Based on Optimized BP Networks

In this paper, an in-depth study and analysis of college English prediction skills and reading efficiency are conducted using an optimized BP network algorithm, which is designed separately, and the accuracy of scoring results is verified and the experimental results are analyzed using two datasets; finally, under the guidance of writing feedback theory and data visualization design criteria, a reasonable visualizable writing feedback is created. The learning and training of the BP neural network model revealed that there was significant information loss in the similar nearest-neighbor user dataset used for training, and there were instances where some items were rated by the target user but not by similar nearest-neighbor users. As a result, such data is useless in training, and the neural network loses a significant amount of useful information when learning. To address this issue, this paper proposes using the singular value decomposition technique for filling, which alleviates the sparsity of the filled matrix data and improves the accuracy of recommendations even more. Both scoring models constructed using 1D CNN and LSTM networks belong to the second class of models, and this type of “end-to-end” scoring model does not require feature engineering. Finally, using 650 spoken recordings and their corresponding manual scoring data, the model is trained and tested. The experimental results show that, with a smaller training dataset, the BP network model achieves a better overall scoring performance.

Identifying Intrinsically Disordered Protein Regions through a Deep Neural Network with Three Novel Sequence Features.

The fast, reliable, and accurate identification of IDPRs is essential, as in recent years it has come to be recognized more and more that IDPRs have a wide impact on many important physiological processes, such as molecular recognition and molecular assembly, the regulation of transcription and translation, protein phosphorylation, cellular signal transduction, etc. For the sake of cost-effectiveness, it is imperative to develop computational approaches for identifying IDPRs. In this study, a deep neural structure where a variant VGG19 is situated between two MLP networks is developed for identifying IDPRs. Furthermore, for the first time, three novel sequence features—i.e., persistent entropy and the probabilities associated with two and three consecutive amino acids of the protein sequence—are introduced for identifying IDPRs. The simulation results show that our neural structure either performs considerably better than other known methods or, when relying on a much smaller training set, attains a similar performance. Our deep neural structure, which exploits the VGG19 structure, is effective for identifying IDPRs. Furthermore, three novel sequence features—i.e., the persistent entropy and the probabilities associated with two and three consecutive amino acids of the protein sequence—could be used as valuable sequence features in the further development of identifying IDPRs.

Development of a prototype service for the diagnosis of diabetic retinopathy based on fundus photos using artificial intelligence methods

Justification and purpose of the study. The high social significance of diabetic retinopathy (DR), the complexity of early diagnosis and monitoring of the disease determine the urgency of developing a system for diagnosing diabetic changes in the fundus using artificial intelligence (AI) methods. The aim of the work is to build a demo prototype of a WEB service for recognizing signs of DR from fundus images using machine learning methods using the Python language and the Django framework.Materials and methods. The study used the Messidor dataset (1200 eyes), which is publicly available on the Internet and includes photos of healthy fundus (546 eyes) and fundus with pathology (654 eyes). With the help of the augmentation method, this set for the study is increased several times. The fundus image recognition system is based on the trained neural network ResNet50. The web service with a connected neural network model was developed on the Django framework.Results. The main result of the research is the development of a test prototype of a service for the diagnosis of diabetic fundus changes based on photos using machine learning tools, demonstrating the great potential of using AI to improve the effectiveness of decisions. The sensitivity of the neural network model during the diagnosis of DR, even on a small test sample and limited training time, was 85 %.Conclusion. The high efficiency and potential of AI methods in the construction of a system for automatic detection of fundus pathology in the framework of the developed in the Helmholtz National Medical Research Center of Eye Diseases automated medical decision-making system. In the future, this service can be used to improve the effectiveness of early diagnosis and monitoring of diabetic changes in the fundus in conditions of reduced availability of primary ophthalmological care in parts of the Russian Federation, including at the pre-medical stage.

Explaining multivariate molecular diagnostic tests via Shapley values

BackgroundMachine learning (ML) can be an effective tool to extract information from attribute-rich molecular datasets for the generation of molecular diagnostic tests. However, the way in which the resulting scores or classifications are produced from the input data may not be transparent. Algorithmic explainability or interpretability has become a focus of ML research. Shapley values, first introduced in game theory, can provide explanations of the result generated from a specific set of input data by a complex ML algorithm.MethodsFor a multivariate molecular diagnostic test in clinical use (the VeriStrat® test), we calculate and discuss the interpretation of exact Shapley values. We also employ some standard approximation techniques for Shapley value computation (local interpretable model-agnostic explanation (LIME) and Shapley Additive Explanations (SHAP) based methods) and compare the results with exact Shapley values.ResultsExact Shapley values calculated for data collected from a cohort of 256 patients showed that the relative importance of attributes for test classification varied by sample. While all eight features used in the VeriStrat® test contributed equally to classification for some samples, other samples showed more complex patterns of attribute importance for classification generation. Exact Shapley values and Shapley-based interaction metrics were able to provide interpretable classification explanations at the sample or patient level, while patient subgroups could be defined by comparing Shapley value profiles between patients. LIME and SHAP approximation approaches, even those seeking to include correlations between attributes, produced results that were quantitatively and, in some cases qualitatively, different from the exact Shapley values.ConclusionsShapley values can be used to determine the relative importance of input attributes to the result generated by a multivariate molecular diagnostic test for an individual sample or patient. Patient subgroups defined by Shapley value profiles may motivate translational research. However, correlations inherent in molecular data and the typically small ML training sets available for molecular diagnostic test development may cause some approximation methods to produce approximate Shapley values that differ both qualitatively and quantitatively from exact Shapley values. Hence, caution is advised when using approximate methods to evaluate Shapley explanations of the results of molecular diagnostic tests.

Optimal design of noise reduction and shape modification for traction gears of EMU based on improved BP neural network

Under high-speed operating conditions, the noise caused by the vibration of the traction gear transmission system of the Electric Multiple Units (EMU) will distinctly reduce the comfort of passengers. Therefore, analyzing the dynamic characteristics of traction gears and reducing noise from the root cause through comprehensive modification of gear pairs have become a hot research topic. Taking the G301 traction gear transmission system of the CRH380A high-speed EMU as the research object and then using Romax software to establish a parametric modification model of the gear transmission system, through dynamics, modal and Noise Vibration Harshness (NVH) simulation analysis, the law of howling noise of gear pair changes with modification parameters is studied. In the small sample training environment, the noise prediction model is constructed based on the priority weighted Back Propagation (BP) neural network of small noise samples. Taking the minimum noise of high-speed EMU traction gear transmission as the optimization goal, the simulated annealing (SA) algorithm is introduced to solve the model, and the optimal combination of modification parameters and noise data is obtained. The results show that the prediction accuracy of the prediction model is as high as 98.9%, and it can realize noise prediction under any combination of modification parameters. The optimal modification parameter combination obtained by solving the model through the SA algorithm is imported into the traction gear transmission system model. The vibration acceleration level obtained by the simulation is 89.647 dB, and the amplitude of the vibration acceleration level is reduced by 25%. It is verified that this modification optimization design can effectively reduce the gear transmission.

The advanced machine learner XGBoost did not reduce prehospital trauma mistriage compared with logistic regression: a simulation study

BackgroundAccurate prehospital trauma triage is crucial for identifying critically injured patients and determining the level of care. In the prehospital setting, time and data are often scarce, limiting the complexity of triage models. The aim of this study was to assess whether, compared with logistic regression, the advanced machine learner XGBoost (eXtreme Gradient Boosting) is associated with reduced prehospital trauma mistriage.MethodsWe conducted a simulation study based on data from the US National Trauma Data Bank (NTDB) and the Swedish Trauma Registry (SweTrau). We used categorized systolic blood pressure, respiratory rate, Glasgow Coma Scale and age as our predictors. The outcome was the difference in under- and overtriage rates between the models for different training dataset sizes.ResultsWe used data from 813,567 patients in the NTDB and 30,577 patients in SweTrau. In SweTrau, the smallest training set of 10 events per free parameter was sufficient for model development. XGBoost achieved undertriage rates in the range of 0.314–0.324 with corresponding overtriage rates of 0.319–0.322. Logistic regression achieved undertriage rates ranging from 0.312 to 0.321 with associated overtriage rates ranging from 0.321 to 0.323. In NTDB, XGBoost required the largest training set size of 1000 events per free parameter to achieve robust results, whereas logistic regression achieved stable performance from a training set size of 25 events per free parameter. For the training set size of 1000 events per free parameter, XGBoost obtained an undertriage rate of 0.406 with an overtriage of 0.463. For logistic regression, the corresponding undertriage was 0.395 with an overtriage of 0.468.ConclusionThe under- and overtriage rates associated with the advanced machine learner XGBoost were similar to the rates associated with logistic regression regardless of sample size, but XGBoost required larger training sets to obtain robust results. We do not recommend using XGBoost over logistic regression in this context when predictors are few and categorical.