Prediction Accuracy Of Model Research Articles

Machine learning and numerical simulation research on specific energy consumption for gradated coarse particle two-phase flow in inclined pipes

In deep-sea mining engineering, accurately predicting the energy required per unit length of pipeline to transport a unit mass of solids (dimensionless specific energy consumption, DSEC) is crucial for ensuring energy conservation and efficiency in the project. Based on our previous work, we utilized the machine learning (ML) and the computational fluid dynamics (CFD)–discrete element method (DEM) method to study the transport characteristics and flow field variations of gradated coarse particles in inclined pipes (gradated particles refer to solid particles mixed in specific size and quantity ratios). First, we collect 1185 sets of data from 13 experimental literature, and after analyzing and processing them, an ensemble model based on four other ML models is developed. Both for pure substance particles (PS) and mixed particles (MP), the prediction accuracy of this ensemble model is relatively higher (PSs are spherical particles with uniform size and density, and MPs are particles with different shapes, sizes, and densities). Then, the CFD-DEM process and the operating conditions include low flow velocity with low volume concentration (2 m/s and 2.5%), low flow velocity with high volume concentration (2 m/s and 7.5%), and high flow velocity with low volume concentration (4 m/s and 2.5%). Under conditions of low flow velocity and low concentrations, as well as high flow velocity and low concentrations, the DSEC hardly changes with the variation of the pipe inclination angle. Under low flow velocity and high-concentration conditions, as the pipe gradually becomes vertical, the value of DSEC gradually increases.

Assessment of Multifidelity Surrogate Approaches for Expedient Loads Prediction in High-Speed Flows

Fast and accurate predictions about the flowfield surrounding a hypersonic flight vehicle are necessary for both early design phases and autonomous vehicle control. Data-driven model reduction presents an opportunity to harness the accuracy of high-fidelity techniques for expedient online application. Surrogate modeling is one approach that seeks to emulate the response of a system with evaluation times much faster than the data source. Multifidelity surrogate modeling seeks to reduce the amount of high-fidelity data needed for a sufficiently accurate online model. The objective of this paper is to assesses the tradeoffs between a kriging-based multifidelity model and a neural-network-based multifidelity model, applied to high-speed flow past a canonical flight geometry. All models studied in this work reproduce the results of the benchmark simulations with less than 100 training cases. The kriging models marginally outperform the neural-network-based models in accuracy, but requirements for training and storing the kriging models scale exponentially as the training data gets larger. Kriging models offer built-in uncertainty quantification and require significantly less decision making about hyperparameters and architecture choices. This study also indicates that the inclusion of data from classical engineering methods can improve the prediction accuracy of surrogate models at practically no cost.

A novel classification algorithm for customer churn prediction based on hybrid Ensemble-Fusion model

Nowadays, customer churn issues are becoming more and more important, which is one of the most important metrics for evaluating the health of a business it is difficult to measure success without measuring customer churn metrics. However, it has become a challenge for the industry to predict when customers are churning or preparing to churn and to take the necessary action at the critical time before they do. At the same time, how to keep the place of deep research on the 17 machine learning algorithms in 9 major classes of machine learning classics production is the first problem we are facing. Through customer churn deep research, we mentioned the Ensemble-Fusion model based on machine learning and introduced a smart intelligent system to help reduce the actual customer churn about the production. Comparing with most popular predictive models, such as the Support vector machine algorithm, Random Forest algorithm, K-Nearest-Neighbor algorithm, Gradient boosting algorithm, Logistic regression algorithm, Bayesian algorithm, Decision tree algorithm, and Neural network algorithm are applied to check the effect on accuracy, AUC, and F1-score. By comparing with 17 algorithms in 9 categories of machine learning classics, the data prediction accuracy of the Ensemble-Fusion model reaches 95.35%, AUC score reaches 91% and F1-Score reaches 96.96%. The experimental results show that the data prediction accuracy of the Ensemble-Fusion model outperforms that of other benchmark algorithms.

Machine learning can be as good as maximum likelihood when reconstructing phylogenetic trees and determining the best evolutionary model on four taxon alignments

Phylogenetic tree reconstruction with molecular data is important in many fields of life science research. The gold standard in this discipline is the phylogenetic tree reconstruction based on the Maximum Likelihood method. In this study, we present neural networks to predict the best model of sequence evolution and the correct topology for four sequence alignments of nucleotide or amino acid sequence data. We trained neural networks with different architectures using simulated alignments for a wide range of evolutionary models, model parameters and branch lengths. By comparing the accuracy of model and topology prediction of the trained neural networks with Maximum Likelihood and Neighbour Joining methods, we show that for quartet trees, the neural network classifier outperforms the Neighbour Joining method and is in most cases as good as the Maximum Likelihood method to infer the best model of sequence evolution and the best tree topology. These results are consistent for nucleotide and amino acid sequence data. We also show that our method is superior for model selection than previously published methods based on convolutionary networks. Furthermore, we found that neural network classifiers are much faster than the IQ-TREE implementation of the Maximum Likelihood method. Our results show that neural networks could become a true competitor for the Maximum Likelihood method in phylogenetic reconstructions.

Peridynamics-fueled convolutional neural network for predicting mechanical constitutive behaviors of fiber reinforced composites

Despite advancements in predicting the constitutive relationships of composite materials, characterizing the effects of microstructural randomness on their mechanical behaviors remains challenging. In this study, we propose a data-driven convolutional neural network (CNN) to efficiently predict the stress-strain curves containing three key material features (Tensile strength, modulus, and toughness) of fiber reinforced composites. Firstly, stress-strain curves for composites with arbitrary fiber distributions were generated using experimentally validated peridynamics (PD) model. Principal component analysis (PCA) was then employed to learn these curves in a lower-dimensional space, reducing computational costs. Subsequently, these reduced data, along with randomly distributed microstructural features, were used to train, validate, and evaluate the CNN models. The combined CNN and PCA model accurately predicted stress-strain curves with maximum errors of 2.5 % for tensile strength, 10% for modulus, and 20 % for toughness. Furthermore, data augmentation and Mean Squared Error (MSE) as a loss function significantly enhanced the model's prediction accuracy. Our findings indicated that DenseNet121 outperformed other CNN models in predicting the properties of fiber-reinforced materials, further demonstrating the effectiveness of the proposed model. This work successfully demonstrates the applicability of a data-driven CNN approach to predict stress-strain relations for engineering materials with intricate heterogeneous microstructures, paving the way for data-driven computational mechanics applied in composites.

A three-directional stress-strain model-based physics-embedded prediction framework for metal tube full-bent cross-sectional characteristics

A metal tube system is known as the industrial blood vessel, among which the bent section is the most vulnerable part. The cross-sectional defects (CSDs) of the bent tube cause the flow fluctuation of the fluid inside the tube. The existing defect characterization methods are roughly presented by describing CSDs in some specific cross-sections, which results in the lack of the tube full-bent section (FBS) characteristic information. To comprehensively describe and predict the tube FBS characteristics, an advanced physics-embedded CSDs prediction framework is proposed. This framework includes an FBS-neutral layer displacement angle (NLDA) prediction module and an FBS-CSDs prediction module, which uses the method that integrates the analytical model and BiLSTM-based deep learning (DL) models to predict the CSDs in the FBS of the tube. A novel analytical model of CSDs that considers both three-directional stresses and strains during tube bending is embedded in the FBS-CSDs prediction module. The analytical model provides the initial predicted values of CSDs through the NLDA sequence obtained from the FBS-NLDA module. The inaccurate CSDs are then treated as physical information to be fed into DL models for further correction and prediction. The prediction performance of this framework is validated through numerical simulations and experiments. The results prove that the framework can accurately predict the CSDs in the tube FBS. The integration of DL models with the analytical model not only overcomes the limitations of the analytical model, but also improves the prediction accuracy and convergence speed of DL models.

A new framework for interval wavelength selection based on wavelength importance clustering

BackgroundWavelength selection is one of the key steps in spectral analysis and plays an irreplaceable role in improving model prediction accuracy and computational efficiency. High-dimensional spectral datasets contain substantial irrelevant information and redundant variables. Whereas, at current stage, such problem can be solved by existing abundant wavelength selection methods. However, it is difficult to achieve the balance between strong wavelength interpretability and prediction accuracy by those methods. As a result, there is an urgent need for a new method that can reach the point of balance. Resultswe propose a new framework for wavelength selection based on wavelength importance clustering (WIC) which attempts to establish a hierarchical relationship between wavelength points and attributions of response through a clustering algorithm, consequently, performing combinatorial and filtering to obtain the optimal wavelength combinations. In this paper, a new wavelength selection method (WIC-WRCKF) is constructed based on WIC, and four commonly used wavelength selection methods are selected to be compared with WIC-WRCKF. A large number of experiments are carried out on three publicly available datasets as well, namely, wheat, corn, and tablets. Compared with other methods, WIC-WRCKF has the highest prediction accuracy with high stability on the three datasets, and the number of wavelengths selected is small and highly interpretative, indicating that WIC-WRCKF has a better predictive ability. SignificanceThe wavelength selection method can significantly improve the model prediction accuracy, and the WIC architecture can effectively exploit the essence of the spectral data, which has great potential in the application of wavelength selection.

Root phenotype detection of rice seedling under nitrogen conditions based on terahertz imaging technique

Root phenotype detection is the basis for screening of dominant root and improving of seed breeding. The efficiency of root phenotype detection is restricted by the concealment and complexity of crop roots. In this paper, a new method based on terahertz imaging technology was proposed to detect the phenotypic characteristics of rice root and quantitatively predict root nitrogen content. First, the terahertz imaging spectral data of rice root were preprocessed with de-overlapping, reconstruction, enhancement, and segmentation. Then, the phenotypic characteristics of root length, diameter and surface area of rice roots were extracted according to the refinement algorithm. In addition, three linear regression models were fitted between root phenotype and root nitrogen content. Finally, Convolutional Neural Network (CNN), Genetic Algorithm-Back Propagation Neural Network (GA-BPNN), and Sparrow Search Algorithm-Support Vector Regression (SSA-SVR) were used to predict nitrogen content in rice roots by training and testing terahertz time domain data. Compared with two kinds of root phenotypic analysis software, the average errors of root length, diameter and root surface area calculated in this paper were 10.68 %, 7.29 % and 3.49 %, respectively. The fitting determination coefficients between root length, root surface area and the root nitrogen content were 0.88 and 0.87 after removing three groups of high-nitrogen data and one group of contrast data. The prediction accuracy of SSA-SVR model was 0.99 and the root mean square error was 0.05. Studies show that terahertz imaging technique can provide an effective analytical way for qualitative analysis of root phenotype.

The surrogate model for short-term extreme response prediction based on ANN and Kriging algorithm

Numerical simulation is a common method for calculating the short-term extreme response of floating offshore wind turbines (FOWTs). However, it requires significant computational resources. This study presents a dynamic response database for a 5MW semi-submersible FOWT under complex environmental conditions, including wind speed, effective wave height, and wave spectral peak period, using a numerical model. The peak over threshold (POT) method can be used to obtain the parametric database of short-term extreme responses, which includes the short-term extreme response distribution parameters for four responses: float surge, mooring tension, outward bending moment at the leaf root surface (OoPBM) and tower base pitching moment (TBPM). And the parameter database is applied to train models such as the Genetic Algorithm optimization Back Propagation neural network (GA-BP) and Kriging algorithm models. The research indicates that a correlation can be established between environmental conditions and short-term extreme response parameters using two algorithms. The accuracy of surrogate model prediction for some parameters can be improved by grouping the data based on wind speed and training separately. Additionally, selecting the appropriate surrogate model for each parameter separately can improve the accuracy of short-term extreme response prediction.

Development of integrated factor modeling: Inhibiting Ganoderma lignocellulosic enzymes while promoting Trichoderma sporulation for enhanced plant disease control

Ganoderma, a fungal pathogen, is the causal agent for basal stem rot disease in oil palms, which has led to significant losses in palm oil production. The fungal production of lignocellulosic enzymes is crucial for successful infection, hence the disease severity could be significantly reduced by inhibiting these enzymes. The current study investigated the impact of integrated factors on inhibiting these enzymes and promoting Trichoderma virens spore production using response surface methodology. The identified optimal conditions of temperature (27.5–36.5 °C), boron (0.07–0.30 mg/mL), potassium (3.5 mg/mL), and mancozeb (1.0 mg/mL) can lead to the production of ≥5.0✕105 spores/mL of T. virens, while minimizing the laccase, cellulase, and xylanase activities to ≤0.08, 0.0, and 0.0 U/mL, respectively. The model's prediction accuracy was 81.25 %. In a greenhouse application, soil temperature was increased to 43.7 °C by solarization before applying the optimal concentration of boron, potassium, mancozeb, and T. virens. The results demonstrated the inhibition of lignocellulosic enzymes in G. boninense, as indicated by 69.0 % and 68.7 % lignocellulose contents in Trichoderma-treated and Ganoderma + Trichoderma-treated oil palm roots, respectively, which were higher than that in Ganoderma-treated roots (58.1 %). This highlights the practical relevance of our findings. Therefore, these results will help the oil palm growers to develop disease management strategies to control basal stem rot.

A three-stage wavelength selection algorithm for near-infrared spectroscopy calibration

The near-infrared spectral data is highly high dimensional and contains redundant information, it is necessary to identify the most representative characteristic wavelengths before modeling to improve model accuracy and reliability. At present, there are many methods for selecting the characteristic wavelengths of NIR spectroscopy, but the collinearity among wavelengths is still a main issue that leads to poor model effects. Therefore, this study proposes a three-stage wavelength selection algorithm (Stage III) to reduce redundancy in NIR spectral data and collinearity between wavelength variables, resulting in a simpler and more accurate predictive model. The research uses a public NIR data set of corn samples as its subject. Initially, the wavelengths with the higher correlation coefficients are chosen after calculating the relationship coefficients between every wavelength vector and the concentration vector. On this basis, the correlation coefficients between the vectors of each wavelength point are calculated, and those wavelength points with smaller correlation coefficients with other wavelength points are selected. Ultimately, the stepwise regression analysis selects the wavelengths that provide substantial value to the model as the variables for modeling, leading to the development of a multiple linear regression model. The results show that the model using the three-stage wavelength selection algorithm outperforms those using the full spectrum, Stages I and Stage II, and the coefficient of determination of the test set of the Stage III-MLR model achieved an accuracy of 0.9360. Instead of the successive projections algorithm (SPA), uninformative variable elimination (UVE), and competitive adaptive reweighted sampling (CARS), Stage III is better in the model prediction accuracy. Therefore, the three-stage wavelength selection algorithm is an effective wavelength selection algorithm that can effectively model NIR spectroscopy, reduce the collinearity between the wavelength variables, simplify the complexity of the model, and improve the prediction precision of the model.

Measure-correlate-predict methods to improve the assessment of wind and wave energy availability at a semi-exposed coastal area

To reduce the carbon footprint of near-shore and land-based Recirculating Aquaculture Systems (RAS), there is interest to harness renewable energy from local sources. When estimating power availability from wind and waves, it is imperative to ensure that the metocean resource data accurately represents the energy flux at the specific site. Here, four measure-correlate-predict (MCP) models are investigated for wind speed, significant wave height and peak wave period. The MCP relationships are obtained from higher spatial resolution wind and wave data, from a weather research and forecasting model, met-mast measurements at Stord airport (wind), and spectral downscaled DNORA/SWAN model (wave), correlated to the NORA3 database over a common time interval. These relations are used to improve the predictions of metocean parameters from the NORA3 model at the RAS location for the period 2012–2021, and hence allow more accurate estimates of renewable power availability. The prediction accuracy of the MCP models improves in all four cases when an appropriate data sorting criterion and algorithm is implemented. Between 2012 and 2021, average wind power density and wave power density are lowest in July at 182 W/m2 and 0.6 kW/m, and highest in December at 538 W/m2 and 7.5 kW/m, respectively.

Batch subproblem coevolution with gaussian process-driven linear models for expensive multi-objective optimization

The efficacy of surrogate-assisted multi-objective evolutionary algorithms (SAMOEAs) in addressing expensive multi-objective optimization problems (MOPs) is contingent upon the modeling techniques and model-based infill sampling strategies. In addressing this pivotal aspect, this paper introduces a pioneering methodology known as batch subproblem coevolution with Gaussian process-driven linear models (BSCo-GPLM). Specifically, from a modeling perspective, BSCo-GPLM decomposes the MOP into single-objective subproblems. Following this decomposition, for each subproblem, a Gaussian process-driven linear model (GPLM) is collaboratively trained to prevent overfitting and improve prediction accuracy. Regarding infill sampling, collaborative optimization of all GPLMs yields optimal candidate solutions for each subproblem, organized into coherent clusters. Within each cluster, only the solution with the highest utility is evaluated. Relying on the heightened prediction accuracy of the GPLM model and an efficient batch sampling strategy, BSCo-GPLM exhibits clear superiority over state-of-the-art SAMOEAs in effectively addressing expensive MOPs. The source code of BSCo-GPLM is available at https://github.com/CIAM-Group/EvolutionaryAlgorithm_Codes/tree/main/BSCo-GPLM.

An accident severity prediction framework with consider features interaction

Accident severity prediction is increasingly important for preventing and reducing traffic accident losses. However, many studies always ignore the complex relationship between features during feature selection. To improve the prediction accuracy of the accident severity prediction model, this paper considers the interactions between multiple features. First, the feature selection algorithm of Recursive Feature Elimination with Cross-Validation (RFECV) is improved by using Shapley Additive explanations (SHAP) as the feature importance assessment metric. Then, to decrease the time expense of manually finding hyperparameters of the model, the Hunter Prey Optimization (HPO) algorithm is introduced and logistic mapping together with stochastic perturbation is added to it, which makes it easier to skip out of the partial optimum during the optimization search. Finally, the improved HPO algorithm is used to optimize the hyperparameters of the CatBoost model. The US traffic accident dataset is introduced for the validity of the proposed framework. Experimental results show that the proposed framework achieves a prediction accuracy of 96.63%, which is better than other state-of-the-art methods. The high accuracy of the prediction model can help decision-makers develop more rational transportation policies, and this study also proposes some traffic management measures based on the selected features.

Phenotyping Alfalfa (Medicago sativa L.) Root Structure Architecture via Integrating Confident Machine Learning with ResNet-18.

Background: Root system architecture (RSA) is of growing interest in implementing plant improvements with belowground root traits. Modern computing technology applied to images offers new pathways forward to plant trait improvements and selection through RSA analysis (using images to discern/classify root types and traits). However, a major stumbling block to image-based RSA phenotyping is image label noise, which reduces the accuracies of models that take images as direct inputs. To address the label noise problem, this study utilized an artificial intelligence model capable of classifying the RSA of alfalfa (Medicago sativa L.) directly from images and coupled it with downstream label improvement methods. Images were compared with different model outputs with manual root classifications, and confident machine learning (CL) and reactive machine learning (RL) methods were tested to minimize the effects of subjective labeling to improve labeling and prediction accuracies. Results: The CL algorithm modestly improved the Random Forest model's overall prediction accuracy of the Minnesota dataset (1%) while larger gains in accuracy were observed with the ResNet-18 model results. The ResNet-18 cross-population prediction accuracy was improved (~8% to 13%) with CL compared to the original/preprocessed datasets. Training and testing data combinations with the highest accuracies (86%) resulted from the CL- and/or RL-corrected datasets for predicting taproot RSAs. Similarly, the highest accuracies achieved for the intermediate RSA class resulted from corrected data combinations. The highest overall accuracy (~75%) using the ResNet-18 model involved CL on a pooled dataset containing images from both sample locations. Conclusions: ResNet-18 DNN prediction accuracies of alfalfa RSA image labels are increased when CL and RL are employed. By increasing the dataset to reduce overfitting while concurrently finding and correcting image label errors, it is demonstrated here that accuracy increases by as much as ~11% to 13% can be achieved with semi-automated, computer-assisted preprocessing and data cleaning (CL/RL).

Prediction for Global Whipping Responses of a Large Cruise Ship Under Unprecedented Sea Conditions Using an LSTM-Based Encoder–Decoder Model

Abstract Global whipping responses contribute to a significant increase in vertical bending moments (VBM), making their accurate prediction crucial for ship safety. In this study, a long short-term memory (LSTM)-based encoder–decoder model is established to predict the whipping responses under varying sea states. The model is trained on a comprehensive dataset, which includes motion data and VBM history of a cruise ship under various sea conditions. This dataset is established via numerical simulation, ensuring a wide range of scenarios for the model to learn from. The efficacy of the LSTM encoder–decoder model in capturing global whipping responses is initially verified under a single sea condition case. This step confirms the model's ability to accurately predict vertical bending moments under known conditions. Subsequently, the model's performance under unprecedented sea conditions is examined. Given that the distribution of training data significantly influences the model's performance and the data from diverse sea conditions typically exhibit distinct data distribution, a mixed data training strategy is employed during the training process in this scenario. The results indicate that the LSTM encoder–decoder model effectively captures whipping responses. Furthermore, the mixed data training strategy significantly improves the model's prediction accuracy for global whipping responses under unprecedented sea conditions.

Construction of a full color gamut color mixture model and a color prediction neural network model for rotor spinning melange yarn

This research is devoted to solving the bottleneck problem of spinning and color prediction of rotor spun melange yarn within the full color gamut range. Firstly, a full color gamut grid color mixture model with hue regulation range of 0–360°, saturation regulation range of 0-1 and lightness regulation range of 0-1 was constructed based on the preferred seven primary-color fibers through grid color mixing. Secondly, in the full color gamut grid color mixture model, 55 representative grid points were planned and corresponding melange yarns and fabrics were prepared based on grid point parameters, which served as the training set, testing set, and validation set for constructing neural network prediction models. Thirdly, based on the demand for color prediction of melange yarns, a color prediction model and a blending ratio prediction model with six sub-models are constructed by taking the reflectivity of the melange yarn fabric and the blending ratio of the primary-color fibers as input and output values. Finally, six prediction samples were produced to verify the predictive performance of the neural network prediction model. The results show that the average color difference between the predicted and measured colors is 1.06, and the average root mean square error value between the predicted and actual blending ratios is 0.48%. It indicates that the neural network prediction model based on the grid mixing of seven primary-color fibers can adapt to the prediction of the mixing color and blending ratio of primary-color fibers in the full color gamut. Due to the construction of six sub-models based on local hue regions, the prediction accuracy of the neural network prediction model has been greatly improved.

EXPRESS: Joint Assortment and Price Optimization with Multiple Purchases

In this paper, we incorporate the effects of multi-unit purchases into consumer choice models and investigate the associated joint assortment and pricing problems. Under the proposed two-stage choice framework, consumers first form a consideration set, and then select a product with the optimal quantity from the consideration set to maximize their utility. For the joint assortment and pricing problem with homogeneous consumers, we find that a multi-purchase-willingness-ordered assortment is optimal under certain technical conditions, and propose a polynomial-time algorithm to find the optimal assortment and prices for general cases. We further show that the optimal prices and assortment size are not monotone in multi-purchase preference and multi-purchase resistance. For the joint assortment and pricing problem with heterogeneous consumers, we first prove that this problem is NP-hard. We then consider a scenario with discrete prices and we develop a fully polynomial-time approximation scheme by considering a finite number of consumer segments and nested consideration sets. We conduct an empirical study on the JD.com dataset and show that the new choice framework can improve model fitting and prediction accuracy compared with existing choice models with multiple purchases. We reveal that our model is especially suitable for datasets exhibiting a strong diminishing marginal effect for product variety. We further extend our analysis to price competition and establish the existence and uniqueness of a Nash equilibrium. For the joint optimization with fixed costs, we develop an algorithm that can find a solution guaranteeing at least half the optimal expected profit.

Predicting the Population Size and Potential Habitat Distribution of Moschus berezovskii in Chongqing Based on the MaxEnt Model

The forest musk deer (Moschus berezovskii) is a national Class I protected wild animal in China, and the IUCN Red list classifies it as globally endangered. It has significant value in traditional Chinese medicine and spices. However, wild M. berezovskii has faced a severe population decline due to human hunting, habitat loss, and fragmentation. Thus, studying its population size and distribution pattern is of great importance to develop effective conservation measures. Here, we determined the optimal MaxEnt model and used stratified sampling and the fecal pile counting method to predict the population size and potential habitat distribution of wild M. berezovskii in Chongqing using 133 species distribution points and 28 environmental variables. The results were as follows: (1) When the optimal model parameters were RM = 3.5 and FC = LQHPT, it had high model prediction accuracy (AUC = 0.909 ± 0.010, TSS = 0.663). (2) Under various climatic, topographic, vegetation, and anthropogenic disturbance scenarios, M. berezovskii was primarily distributed in northern, eastern, southwestern regions of Chongqing, covering an area of approximately 5562.80 km2. (3) The key environmental factors affecting the potential habitat distribution of M. berezovskii were elevation (36.5%), normalized difference vegetation index (NDVI, 16.6%), slope (11.8%), and land-use type (7.6%), whereas climate and anthropogenic disturbance factors had relatively little influence. (4) A population estimation for M. berezovskii identified approximately 928 ± 109 individuals in Chongqing. We recommend prioritizing the preservation of high-altitude habitats and native vegetation to mitigate human interference and minimize road damage. In summary, our results can enhance the understanding of M. berezovskii distribution and provide a basis for effective conservation and management initiatives.

Enhancing Radiology Training with GPT-4: Pilot Analysis of Automated Feedback in Trainee Preliminary Reports

Rationale and ObjectivesRadiology residents often receive limited feedback on preliminary reports issued during independent call. This study aimed to determine if Large Language Models (LLMs) can supplement traditional feedback by identifying missed diagnoses in radiology residents’ preliminary reports. Materials & MethodsA randomly selected subset of 500 (250 train/250 validation) paired preliminary and final reports between 12/17/2022 and 5/22/2023 were extracted and de-identified from our institutional database. The prompts and report text were input into the GPT-4 language model via the GPT-4 API (gpt-4-0314 model version). Iterative prompt tuning was used on a subset of the training/validation sets to direct the model to identify important findings in the final report that were absent in preliminary reports. For testing, a subset of 10 reports with confirmed diagnostic errors were randomly selected. Fourteen residents with on-call experience assessed the LLM-generated discrepancies and completed a survey on their experience using a 5-point Likert scale. ResultsThe model identified 24 unique missed diagnoses across 10 test reports with i% model prediction accuracy as rated by 14 residents. Five additional diagnoses were identified by users, resulting in a model sensitivity of 79.2%. Post-evaluation surveys showed a mean satisfaction rating of 3.50 and perceived accuracy rating of 3.64 out of 5 for LLM-generated feedback. Most respondents (71.4%) favored a combination of LLM-generated and traditional feedback. ConclusionThis pilot study on the use of LLM-generated feedback for radiology resident preliminary reports demonstrated notable accuracy in identifying missed diagnoses and was positively received, highlighting LLMs' potential role in supplementing conventional feedback methods.