Set Of Predictors Research Articles

Intelligent identification of picking periods of Lu’an Guapian tea by an indicator displacement colorimetric sensor array combined with machine learning

Lu’an Gua Pian (LAGP) tea is one of the most famous green teas in China. The quality of green tea is related to its picking periods, especially the green tea before Qingming Festival (usually April 6th) is highly praised as precious in the market. In this work, a simple and cheap indicator displacement colorimetric sensor array combined with smartphone was developed to rapidly identify LAGP picked during different picking periods. First, the chemical component contents of LAGP picked before and after Qingming Festival were analyzed. Second, a well-designed colorimetric sensor array was proposed based on the tea component contents differences. Finally, machine learning was used to process the array data taken by a smartphone. By comparison, the accuracy of the best model for the prediction set was 97%. Meanwhile, the multi-channel advantages of the sensing array were demonstrated by an ablation experiment. In addition, the method achieved an AGREE analysis score of 0.88, indicating that it was environmental-friendly.

Evaluating Synthetic Data Augmentation to Correct for Data Imbalance in Realistic Clinical Prediction Settings.

Predictive modeling holds a large potential in clinical decision-making, yet its effectiveness can be hindered by inherent data imbalances in clinical datasets. This study investigates the utility of synthetic data for improving the performance of predictive modeling on realistic small imbalanced clinical datasets. We compared various synthetic data generation methods including Generative Adversarial Networks, Normalizing Flows, and Variational Autoencoders to the standard baselines for correcting for class underrepresentation on four clinical datasets. Although results show improvement in F1 scores in some cases, even over multiple repetitions, we do not obtain statistically significant evidence that synthetic data generation outperforms standard baselines for correcting for class imbalance. This study challenges common beliefs about the efficacy of synthetic data for data augmentation and highlights the importance of evaluating new complex methods against simple baselines.

Exploring the analytical potential of total reflection X-ray fluorescence (TXRF) combined with partial least square (PLS) for simple determination of ash content in various coal types

Ash content, as a crucial indicator of coal quality, its rapid and accurate determination is pivotal to improve the energy utilization of coal and reduce environmental pollution. Traditional spectroscopic methods face significant challenges in acquiring accurate information from coal samples due to the notorious matrix effects arising from their complex composition, vast molecular structure, and diverse coal types. In this study, the feasibility of total reflection X-ray fluorescence (TXRF) combined with partial least squares (PLS) for the determination of coal ash was firstly investigated based on the TXRF being unaffected by matrix effects. Firstly, coal samples were prepared as suspensions, and the effects of sample particle size and different dispersants on the results of TXRF analyses were evaluated. The accuracy and applicability of the chosen sample preparation strategies were further validated using inductively coupled plasma mass spectrometry (ICP-MS) and two certified reference materials (CRMs). Subsequently, based on the analysis of 19 coal samples, the impact of three different predictive variables on the performance of the PLS model was investigated: (a) TXRF full spectrum normalized by the net intensity of the internal standard; (b) net intensity of characteristic peaks for 12 elements (Al, Si, K, Ca, Ti, Fe, Cr, Mn, Cu, Ni, and Sr) normalized by the net intensity of the internal standard; (c) concentrations of the aforementioned 12 elements. The results demonstrate that the PLS model constructed usingthe TXRF full spectrum normalized by the net intensity of the internal standard exhibits the best predictive capabilities, with the determination coefficient of calibration set (R2) and mean square error (MSE) of the prediction set reaching 0.9736 and 0.99 %, respectively. Moreover, the measurement accuracy of this model was six times greater than that obtained with traditional X-ray fluorescence (XRF). Presented analytical results display the possibilities of combining TXRF with PLS for coal quality evaluation.

Process Optimization for Antarctic Krill (Euphausia superba) Sauce Based on Back Propagation Neural Network Combined with Genetic Algorithm

Using Antarctic krill (Euphausia superba) as the research object, we optimized the process conditions for Antarctic krill sauce (AkS) by including three factors (salt addition, rock sugar addition, and the oil-to-material ratio) and sensory evaluation as response values. The data from the response surface were fed into the back propagation (BP) neural network training, generating a model mapping the process conditions and sensory scores, which were subsequently combined with the genetic algorithm (GA) for global optimization to determine the optimal process for AkS preparation. The results revealed that the response surface model was well suited to the BP neural network training and prediction sets, with correlation values of 0.98 and 0.95, respectively. The fitting prediction effect was obvious for the sensory scoring results of the product. The parameters obtained from the GA’s global optimization search accord with the analytical results of the response surface. The findings demonstrated that combining a BP neural network with a GA can enhance the AkS preparation technique. Under optimal processing conditions, AkS has a high sensory score and protein and carbohydrate contents, moderate fat content, minimal fat oxidation, and non-detectable pathogens, indicating that the AkS in this study was nutritious and safe to consume.

Dysregulated immunologic landscape of the early host response in melioidosis.

Melioidosis, a neglected tropical infection caused by Burkholderia pseudomallei, commonly presents as pneumonia or sepsis with mortality rates up to 50% despite appropriate treatment. A better understanding of the early host immune response to melioidosis may lead to new therapeutic interventions and prognostication strategies to reduce disease burden. Whole blood transcriptomic signatures in 164 patients with melioidosis and in 70 patients with other infections hospitalized in northeastern Thailand enrolled within 24 hours following hospital admission were studied. Key findings were validated in an independent melioidosis cohort. Melioidosis was characterized by upregulation of interferon (IFN) signaling responses compared with other infections. Mortality in melioidosis was associated with excessive inflammation, enrichment of type 2 immune responses, and a dramatic decrease in T cell-mediated immunity compared with survivors. We identified and independently confirmed a 5-gene predictive set classifying fatal melioidosis (validation cohort area under the receiver operating characteristic curve 0.83; 95% CI, 0.67-0.99). This study highlights the intricate balance between innate and adaptive immunity during fatal melioidosis and can inform future precision medicine strategies for targeted therapies and prognostication in this severe infection.

Quantitative analysis of zearalenone in wheat leveraging support vector machine and olfactory visualization technology

Zearalenone (ZEN) is a contaminant found naturally in grains such as wheat that poses a potential threat to human and animal health. This study describes a colorimetric sensor array detection system based on chemically reactive dyes and chemometric algorithms for the determination of ZEN in wheat. Firstly, six chemically responsive dyes were screened to construct a colorimetric sensor array to obtain the response fingerprints of the volatile components of wheat. The RGB differences of the chemically reactive dyes before and after the response were then calculated to obtain the color component data of the image. The maximum information coefficient (MIC) and ReliefF methods were then used to select the best color features. Finally, the Support Vector Machine Regression (SVR) prediction model was developed and validated based on the optimized features on 132 sample sets. The results showed that the SVR model developed on the initial 132 wheat samples had good predictive ability. Its root mean square error (RMSE) on the prediction set was 38.1625 μg∙kg−1 with a correlation coefficient (R) 0.9516. In particular, the SVR model combining 13 color components reduced the RMSE on the prediction set to 30.4484 μg∙kg−1, with an improved R of 0.9695. Olfactory visualization techniques were shown to be effective for the determination of ZEN in wheat. This study is expected to provide a scientifically novel approach for quality stabilization during the storage of wheat or other cereals.

Objective identification of meteorological fronts and climatologies from ERA-Interim and ERA5

Abstract. Meteorological fronts are important due to their associated surface impacts, including extreme precipitation and extreme winds. Objective identification of fronts is therefore of interest in both operational weather prediction and research settings. The aim of this study is to produce a front identification algorithm based on earlier studies that is portable and scalable to different resolution datasets. We have made a number of changes to an earlier objective front identification algorithm, applied these to reanalysis datasets, and present the improvements associated with these changes. First, we show that a change in the order of operations yields smoother fronts with fewer breaks. Next, we propose the selection of the front identification thresholds in terms of climatological quantiles of the threshold fields. This allows for comparison between datasets of differing resolutions. Finally, we include a number of numerical improvements in the implementation of the algorithm and better handling of short fronts, which yield further benefits in the smoothness and number of breaks. This updated version of the algorithm has been made fully portable and scalable to different datasets in order to enable future climatological studies of fronts and their impacts.

Nutritional Quality Analysis and Classification Detection of Buckwheat in Different Harvest Periods.

For buckwheat, the optimal harvest period is difficult to determine-too early or too late a harvest affects the nutritional quality of buckwheat. In this paper, physical and chemical tests are combined with a method using near-infrared spectroscopy nondestructive testing technology to study buckwheat harvest and determine the optimal harvest period. Physical and chemical tests to determine the growth cycle were performed at 83 days, 90 days, 93 days, 96 days, 99 days, and 102 days, in which the buckwheat grain starch, fat, protein, total flavonoid, and total phenol contents were assessed. Spectral images of buckwheat in six different harvest periods were collected using a near-infrared spectral imaging system. Four preprocessing methods (SNV, S-G, DWT, and the normaliz function) and three dimensionality reduction algorithms (IVSO, VCPA, VISSA) were used to process the raw buckwheat spectral data, and the full and eigen spectra were established as a random forest (RF). Random forest (RF) and Least Squares Support Vector Machine (LS-SVM) classification models were used to determine the full and eigen spectra, respectively, and the optimal model for the buckwheat single harvest period was determined and validated. Through physical and chemical tests, it was concluded that the 90-day harvest buckwheat grain protein, fat, and starch contents were the highest, and that the total flavonoid and total phenolic contents were also high. The SNV preprocessing method was the most effective, and the feature bands extracted using the IVSO algorithm were more representative. The IVSO-RF model was the best discriminative model for the classification of buckwheat in different harvest periods, with the correct rates of the training and prediction sets reaching 100% and 96.67%, respectively. When applying the IVSO-RF model to the buckwheat single harvest period to verify the classification, the correct rate of the training set for each harvest period reached 96%, and that of the prediction set reached 100%. Near-infrared spectroscopy combined with the IVSO-RF modeling method for buckwheat harvest period detection is a rapid, nondestructive classification method. When this was combined with physical and chemical analyses, it was determined that a growth cycle of 90 days is the best harvest period for buckwheat. The results of this study can not only improve the quality of buckwheat crops but also be applied to other crops to determine their optimal harvest period.

Hyperspectral imaging system for pre- and post-harvest defect detection in paprika fruit

External defects in paprika fruit (Capsicum annuum, L. var. angulosum) directly impact their storage and marketability, potentially leading to internal decay and altering fruit appearance. However, automatically detecting various paprika defects, including cracks, scars, diseases, and bruises, poses a significant challenge due to variations in defect size and visual similarities among different types of defects. This study proposed a comprehensive, accurate, and rapid defect detection method utilizing hyperspectral imaging (HSI) alongside multivariate analysis and imaging technologies. The reflective surface of fruit poses a primary limitation to efficient defect detection using imaging technologies. Therefore, a polarized HSI visible–near-infrared (VIS-NIR, 397–1000 nm) system was developed, and over 500 paprika fruit with various types of defects were imaged using the developed system. Additionally, fruit were manually bruised in different sizes to ensure spectral data variability and measured every 12 hours for two days. Spectral data (over 4580 spectral data) from the regions of interest (ROIs) of the paprika fruit samples, including manually bruised fruit, were extracted. Subsequently, a partial least squares discrimination analysis (PLS-DA) model was developed to distinguish between normal and defective fruit. Remarkably, the developed model achieved an accuracy of over 99 % in predicting normal and defective fruit. Moreover, the chemical images generated enhance the inspection process. The successive projection algorithm (SPA) and variable importance in projection (VIP) were further utilized for critical band selection, yielding an accuracy of 93 % and 98 % in the prediction set, respectively. Furthermore, a classification algorithm based on analysis of variance (ANOVA) was employed to determine the optimal wavelength band ratios (F-values). This approach yielded an impressive accuracy rate of over 88.8 % and further enhanced the performance of the band ratio by an improved watershed segmentation algorithm (IWSA) to reach an accuracy of 91.3 % in multi-defect detection. These findings offer a practical and efficient solution for the external quality inspection of paprika fruit, particularly in cases with naturally occurring defects. By integrating hyperspectral imaging, PLS-DA, band selection methods, ANOVA-based classification, and IWSA, this comprehensive analysis provides a reliable and practical means for identifying and categorizing defects in paprika fruit, ultimately enhancing product quality and consumer satisfaction.

Improved weighting factor selection method of predictive torque control for PMSM

AbstractIn the traditional predictive torque control strategy, the cost function considers two different control objectives: stator flux and torque. Therefore, it is necessary to reasonably set the weighting factor to ensure that the optimal voltage vector is accurately selected during the control process, thereby achieving better torque and flux control performance. In view of the complex tuning process caused by traditional predictive torque control setting method of weighting factor, a new weighting factor selection method for permanent magnet synchronous motor control system is proposed in this paper. The derivation process is based on the equivalent transformation between the torque–flux prediction and the voltage vector prediction. If both the cost functions of the torque–flux and voltage vector reach their minimum values, the same output voltage vector will be selected. Compared with previous studies, the algorithm proposed in this paper can find the precise location of the optimal weighting factor without changing the form of the cost function, which simplifies the parameter tuning process. Experimental results are also provided to validate the effectiveness of the proposed algorithm.

Assessing the effects of mechanical damage on optical properties of strawberries in the 950−1650 nm range

To investigate the effect mechanism of early damage on optical properties of strawberries, the optical parameters, internal qualities and microstructure parameters of bruised samples and intact ones were acquired and quantitatively analyzed. The results showed that the soluble solids content increased with damage time, while the moisture content decreased. Firmness showed no difference between intact and bruised samples. The cell circularity (Cc) initially decreased, then slightly increased over damage time. The average cell diameter (Davg) increased with longer damage durations, and the average values of cell wall thickness (Tcw) decreased with longer damage durations. The absorption coefficient (µα) values of bruised samples were higher than the intact ones at 970 nm and 1190 nm. For reduced scattering coefficient (µ′s), at 970 nm and 1190 nm, differences were observed between intact strawberries and bruised samples subjected to three different durations of damage. The average R values of µ′s with Cc, Tcw and Davg were 0.79, 0.93 and −0.93, respectively, confirming a strong correlation of µ′s with cell diameter and cell wall thickness. The support vector machine classification model built based on µα spectra had better performance than that based on µ′s spectra, and the model based on combinations of µα and µ′s (µeff) exhibited the highest accuracy for both the calibration set and prediction set with accuracies of 96.00 % and 92.19 %, respectively. The study provides a theoretical basis for detecting latent damaged strawberries based on optical properties.

High-precision detection of dibutyl hydroxytoluene in edible oil via convolutional autoencoder compressed Fourier-transform near-infrared spectroscopy

The quality of edible oils is closely related to their chemical compositions. Antioxidants have widespread application in edible oil production. In this study, a pioneering detection approach involving the use of a one-dimensional convolutional autoencoder (1D-CAE) was introduced to compress spectral data for assessing antioxidant levels in edible oils. Fourier-transform near-infrared (FT-NIR) characterisation of edible oil samples with varying antioxidant concentrations was also conducted. An 1D-CAE model was developed to compress different pre-processed spectra into a condensed representation. These compressed features were then integrated with a support vector machine and partial least squares regression models to establish correlations for each target. The study examined the influence of pre-processing steps and feature engineering methods on near-infrared spectral analysis through independent or combined model analysis. The findings revealed that features derived from the 1D-CAE model demonstrated remarkable repeatability and can be utilised to construct robust detection models. The experimental results showed that the optimal detection model derived based on the 1D-CAE compression features has an average R2, RPD and RMSE of 0.9953, 15.1664 and 1.2035, respectively, on the prediction set. FT-NIR spectroscopy can be used to accurately detect butylated hydroxytoluene in edible oils. Therefore, autoencoders are an effective tool in spectroscopic analysis, offering promising avenues for future research and application.

Quantitative analysis of multicomponent mixtures of rubber and additives based on terahertz time-domain spectroscopy and Krawtchouk moments

This paper presents a quantitative analytical method based on terahertz time-domain spectroscopy and chemometrics, which was used to detect the content of diphenyl-p-phenylenediamine in a five-component mixture comprising tire rubber and additives. We separately extracted the features of the spectra by using principal component analysis and Krawtchouk moments, and then used partial least squares regression and support vector regression to develop quantitative analysis models. The experimental results show that support vector regression outperforms partial least squares regression in the quantitative analysis of multicomponent mixtures, and the Krawtchouk moments method effectively enhances the prediction accuracy of the model. The Krawtchouk moment combined with support vector regression model shows excellent predictive ability. The correlation coefficient and root mean square error of Krawtchouk moment combined with support vector regression model for the calibration set were 0.9749 and 1.7256%, respectively, and for the prediction set were 0.9669 and 1.9777%, respectively. The results demonstrate that terahertz time-domain spectroscopy combined with Krawtchouk moments and support vector regression is an effective method for determining the antioxidant content in rubber and additive compounds.

Uncertainty estimation of machine learning spatial precipitation predictions from satellite data

Abstract Merging satellite and gauge data with machine learning produces high-resolution precipitation datasets, but uncertainty estimates are often missing. We addressed the gap of how to optimally provide such estimates by benchmarking six algorithms, mostly novel even for the more general task of quantifying predictive uncertainty in spatial prediction settings. On 15 years of monthly data from over the contiguous United States (CONUS), we compared quantile regression (QR), quantile regression forests (QRF), generalized random forests (GRF), gradient boosting machines (GBM), light gradient boosting machines (LightGBM), and quantile regression neural networks (QRNN). Their ability to issue predictive precipitation quantiles at nine quantile levels (0.025, 0.050, 0.100, 0.250, 0.500, 0.750, 0.900, 0.950, 0.975), approximating the full probability distribution, was evaluated using quantile scoring functions and the quantile scoring rule. Predictors at a site were nearby values from two satellite precipitation retrievals, namely PERSIANN (Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks) and IMERG (Integrated Multi-satellitE Retrievals), and the site’s elevation. The dependent variable was the monthly mean gauge precipitation. With respect to QR, LightGBM showed improved performance in terms of the quantile scoring rule by 11.10%, also surpassing QRF (7.96%), GRF (7.44%), GBM (4.64%) and QRNN (1.73%). Notably, LightGBM outperformed all random forest variants, the current standard in spatial prediction with machine learning. To conclude, we propose a suite of machine learning algorithms for estimating uncertainty in spatial data prediction, supported with a formal evaluation framework based on scoring functions and scoring rules.

Bayesian calibration and uncertainty quantification of a rate-dependent cohesive zone model for polymer interfaces

This study presents a rate-dependent cohesive zone model for the fracture of polymeric interfaces and performs a Bayesian calibration, an uncertainty quantification, and a sensitivity analysis for the model. The proposed cohesive zone model accounts for both reversible elastic and irreversible rate-dependent separation sliding deformation at the interface. The viscous dissipation due to the irreversible opening at the interface is modeled using elastic-viscoplastic kinematics that incorporates the effects of strain rate. Inverse calibration of parameters for such complex models through trial and error is challenging due to the large number of parameters of the model. Moreover, the calibrated parameter values are often non-unique and uncertain when the available experimental data is limited. To tackle this challenge, we employ a Bayesian calibration approach to identify parameters from experimental data, the resulting parameters significantly enhance the accuracy of the model. To quantify the uncertainty associated with the inverse parameter estimation, a modular Bayesian approach is employed to calibrate the unknown model parameters, accounting for the parameter uncertainty of the cohesive zone model. The advantages of the Bayesian calibration over a deterministic parameter fit are demonstrated. Further, to quantify the model uncertainties, such as incorrect assumptions or missing physics, a discrepancy function is introduced, which significantly improves the model’s prediction. Finally, the total uncertainty of the model is quantified in a predictive setting. A sensitivity analysis is performed to assess how changes in the input variables of the model affect the peak load, facilitating the identification of a concise set of highly influential parameters. The present approach can be used for calibration and uncertainty quantification for other complex computational mechanics models. It should also facilitate the designing of interface materials under uncertainty.

Utility of radiomic features in predicting clinical outcomes in stage II-III pancreatic cancer.

74 Background: We identified computed tomography (CT)-derived radiomic features predictive of tumor progression within three months, then examined their ability to prognosticate overall survival (OS) along with clinical features in pancreatic cancer. We evaluated these features in patients with unresected pancreatic cancer who underwent stereotactic body radiation therapy (SBRT) in sequence with chemotherapy, but not surgery. Methods: In this retrospective study, we examined a cohort of 101 patients with stage II-III pancreatic cancer who underwent SBRT with sequential chemotherapy at a single institution (Stanford Health Care) between 1999-2020. From their pre-SBRT contrast-enhanced CT images with segmented tumors, delineating regions-of-interest, we extracted 900 radiomic (quantitative pixel-level imaging characteristic) features. In the first phase, we identified radiomic features that predicted rapid tumor progression within three months following SBRT. We divided the dataset into a training set (n = 53) for model development and a test set (n = 48) for evaluation. Using logistic regression with the Least Absolute Shrinkage and Selection Operator algorithm for feature selection and classification, we built a binary prediction model on the training set to identify patients at risk of progression within three months of SBRT. To fine-tune parameters, we performed five-fold cross-validation (CV) on the training set, repeating each set of parameters five times. We assessed model performance on the test set using the area under the curve (AUC). We selected the model with the best AUC, generating the predictive radiomic feature set. In the second phase, we conducted univariate and multivariate Cox regression analyses to assess the relationship between OS and individual clinical variables (age, sex, stage, vessel involvement, tumor location, performance status, body mass index, biological equivalent dose of radiation) and the radiomic feature set as high versus low risk. Results: Our cohort consisted of 48 men (mean age, 70 years ± 11 [SD]) and 53 women (mean age, 67 years ± 13 [SD]). From the first phase, 32 textural features comprised the radiomic feature set that best predicted rapid tumor progression, with mean AUCs of 0.852 (CV, n=53) and 0.814 (test, n=48). In the univariate Cox model, only the radiomic feature set was predictive of OS (hazard ratio, HR, 1.724, p=0.011). In the multivariate Cox model, radiomic features and age were significant predictors of OS, with HR of 1.819 (p=0.007) and 1.024 (p=0.024), respectively. Conclusions: CT-derived radiomic features predict rapid tumor progression following SBRT, confer nearly a twofold increase in mortality risk, and, along with patient age, enhance the identification of patients with stage II-III pancreatic cancer with poor OS. [Table: see text]

Detection of early bruises in plum using hyperspectral imaging combination with machine learning algorithm

Early detection of plum bruising is important in the online postharvest quality sorting process. This paper explores the rapid detection of bruises in plum at five stages (1, 3, 6, 24, and 48 h after bruising) using hyperspectral imaging system. Spectral preprocessing was performed using five methods (unit vector normalization, multiplicative scattering correction, standard normal variate, detrending, and baseline). The support vector machine was established to discriminate the spectral samples of healthy and bruised plums at stage 1 on full wavelengths. The results indicated that the spectral data pretreated by multiple scattering correction yielded better results. The characteristic wavelengths of the spectra were selected by six algorithms (random frog, competitive adaptive reweighted sampling, variable combination population analysis, successive projection algorithm, uninformative variable elimination and bootstrapping soft shrinkage). Subsequently, the support vector machine was established at stage 1 based on these selected characteristic wavelengths. Comparing the results of models, the support vector machine based on the characteristic wavelengths selected by competitive adaptive reweighted sampling generated a satisfied effect with an accuracy of 95% for the calibration set and 98% for the prediction set. This model was selected for bruise detection in plums at the residual stage. The characteristic wavelength grayscale images selected based on the model were used to successfully visualize the plum bruise regions in five stages using minimum noise fraction transformation and the principal component analysis method. The results showed that the hyperspectral imaging technique combined with machine learning algorithm could be used to identify plum bruises at different stages. This study contributes to the development of an online detection system for bruises in plum.

Conspiracy Theatre of the Absurd: “Birds Aren’t Real” as Parodic Hypermimesis

ABSTRACT Emerging from an ad hoc effort to counter Trump supporters at a 2017 Women’s March, the Birds Aren’t Real movement has transformed from viral performance art into a Gen Z response to political conspiracy culture. This article explores the campaign as an extension of the kind of postmodern parody that became iconic with shows like The Colbert Report in which comedians engage in absurd reproductions of increasingly absurd political realities. Inspired by the work of cultural theorist Nidesh Lawtoo, we argue that Birds Aren’t Real engages in parodic hypermimesis, or an ironical meta-mimicking of the digitally networked and infective mimetics responsible for the contemporary proliferation of conspiratorial misinformation. After explaining how parodic hypermimesis works, we illustrate its enactment in the Birds Aren’t Real movement, particularly in the way that it relies on doubling to debase conspiracy rhetoric as a predictable set of topoi, cultivates a conspiracy community via intoxicating affect, and derides the mainstream media by baiting outlets with parodic spectacles. Ultimately, this essay ponders whether, in reverberating conspiracy logics, such well-intentioned parody can successfully ascend the cardinal sins it critiques.

A new method for predicting precipitation δ18O distribution based on deep learning and spatio-temporal clustering

ABSTRACT Predicting precipitation δ18O accurately is crucial for understanding water cycles, paleoclimates, and hydrological applications. Yet forecasting its spatio-temporal distribution remains challenging due to complex climate interactions and extreme events. We developed a method combining spatio-temporal clustering and deep learning neural networks to improve multi-site, multi-year precipitation δ18O predictions. Using a comprehensive dataset from 33 German sites (1978–2012), our model considers precipitation δ18O and its controlling factors, including precipitation and temperature distribution. We applied the K-means ++ method for classification and divided data into training and prediction sets. The convolutional neural network (CNN) model extracted spatial features, while the bi-directional long short-term memory (Bi-LSTM) model focused on temporal features. Spatio-temporal clustering using K-means ++ improved forecast accuracy and reduced errors. This study highlights the potential of deep learning and clustering techniques for forecasting complex spatio-temporal data and offers insights for future research on isotope distributions.

A dielectric method for predicting pear firmness combining deep data augmentation and ensemble learning

Firmness is valuable in evaluating pear quality, as it determines ripeness and storability. This research presents a new method that combines deep data augmentation and ensemble learning for predicting firmness in three pear cultivars using dielectric spectra. Firstly, 126 sets of pear dielectric spectra and firmness data were obtained, and the characteristic dielectric frequencies were selected based on the prediction accuracy of partial least squares regression (PLSR) model established using different preprocessing methods. Then the generative adversarial networks (GAN), least squares GAN (LSGAN), and Wasserstein GAN (WGAN) models with different numbers of generated samples in the model's training set on the prediction accuracy of the PLSR model were compared. Finally, the optimal GAN model was combined with support vector regression (SVR), artificial neural networks (ANN), random forest (RF), and AdaBoost decision tree (AdaBoost-DT) to predict pear firmness. Results showed that the selected characteristic dielectric frequencies included two ε′ and eight ε" points. GAN, LSGAN, and WGAN all improved the prediction accuracy of the PLSR model, and the model accuracy improvement of GAN and LSGAN was proportional to the generated sample numbers added to the training set. GAN had the highest accuracy improvement for PLSR model, with the mean Rp2 (determination coefficient of prediction set) and RMSEP (root mean square error of prediction set) increasing up to 13.70% and −19.48% during 100–1000 epochs (added 126 sets of generated data in the GAN-PLSR model' training set), respectively. Ensemble learning models (RF, AdaBoost-DT) outperformed SVR and ANN. GAN-AdaBoost-DT model added 126 sets of generated data in the training set had the highest prediction accuracy, with average Rp2 and RMSEP of 0.90 and 3.35 N during 300–1000 epochs, respectively. This study proved the feasibility of deep data augmentation and ensemble learning methods to improve the prediction accuracy of pear firmness model without increasing the labor and time costs.