Improved Prediction Model Research Articles

Enhancing creep rupture life prediction of high‐temperature titanium alloys using convolutional neural networks

AbstractPrediction of creep rupture life of high‐temperature titanium alloys is crucial for their practical applications. The efficient representations (features) of the information encoded in the data are essential to achieve an accurate prediction model. Here, using convolutional neural networks (CNN) enhanced features, we obtain largely improved prediction models for creep rupture life. Comparison of CNN‐based features with the original features in describing different samples reveals that the former, by assigning more individualized labels, outperforms the latter and underpins improved prediction models. This work suggests that beyond images, CNN is also suitable for numerical data to obtain enhanced features and surrogate models.

Randomising spatial patterns supports the integration of intraspecific variation in ecological niche models

Ecological niche models (ENMs) are an essential modelling technique in biodiversity prediction and conservation and are frequently used to forecast species responses to global changes. Classic species‐level models may show limitations as they assume species homogeneity, neglecting intraspecific variation. Composite ENMs allow the integration of intraspecific variation by combining intraspecific‐level ENMs, capturing individual environmental responses over the species' geographic range. While recent studies suggest that accounting for intraspecific variation improves model predictions, we currently lack methods to test the significance of the improvement. Here, we propose a null model approach that randomises observed intraspecific structures as an appropriate baseline for comparison. We illustrate this approach by comparing predictive performance of a species‐level ENM to composite ENMs for European beech Fagus sylvatica. To investigate the influence of spatial lineage structure, we tested all models against the same withheld data to allow comparison across models based on five common performance metrics. We found that the species‐level ENM expressed higher overall performance (i.e. AUC, TSS, and Boyce index) and specificity (ability to predict absences), while the composite ENMs achieved higher sensitivity (ability to predict presences). In line with this, the composite ENMs also showed increased sensitivity and decreased specificity compared to the null models that randomised lineage structure. We showed that the assessment of model performance strongly varies based on the used measures, emphasising a careful investigation of multiple measures for evaluation. The application of null models allowed us to disentangle the effect of observed patterns of intraspecific variation in ENMs. Further, we highlight the validation and use of well‐founded subgroups for modelling. Although intraspecific variation improves the prediction of occurrences of European beech, it did not fully outcompete the classic species‐level model and should be used with care and rather to improve understanding and to supplement, not replace, species‐level models.

External Validation of Accelerometry-Based Mechanical Loading Prediction Equations

Accurately predicting physical activity-associated mechanical loading is crucial for developing and monitoring exercise interventions that improve bone health. While accelerometer-based prediction equations offer a promising solution, their external validity across different populations and activity contexts remains unclear. This study aimed to validate existing mechanical loading prediction equations by applying them to a sample and testing conditions distinct from the original validation studies. A convenience sample of 49 adults performed walking, running, and jumping activities on a force plate while wearing accelerometers at their hip, lower back, and ankle. Peak ground reaction force (pGRF) and peak loading rate (pLR) predictions were assessed for accuracy. Substantial variability in prediction accuracy was found, with pLR showing the highest errors. These findings highlight the need to improve prediction models to account for individual biomechanical differences, sensor placement, and high-impact activities. Such refinements are essential for ensuring the models’ reliability in real-world applications, particularly in clinical and biomechanical research contexts, where accurate assessments of mechanical loading are critical for designing rehabilitation programs, injury prevention strategies, and optimizing bone health interventions.

Environmental associations and the paleoecological significance of the genus Pyxidicula Ehrenberg,1838 and Pyxidicula muskegii sp. nov. in the Hudson and James Bay region peatlands, Canada

An environmental proxy routinely used in paleoecological studies of peatlands is testate amoebae, which are used to infer successional and hydrological shifts through time. However, since many species of protists globally remain undescribed, important ecological information about community assemblages and their interaction with environmental conditions in both the past and present has potential gaps. During the analysis of testate amoeba assemblages from peat cores across the Hudson and James Bay region of Canada, a testate amoeba that has not been formally described was discovered with high abundance in subfossil assemblages (up to 50%). We describe this new species, Pyxidicula muskegii sp. nov., using morphometric measurements of tests from these peat core records and demonstrate its environmental preferences using modern samples collected from the same region. The hemispheric shape and irregular areolar surface pattern on the test, along with its preference for lawns in both fens and bogs, support that P. muskegii is a new species. This species is also found in much lower relative abundance in modern samples (<3%) than in paleo-peat core records. This difference in modern and paleo-assemblage suggests that there is a taphonomic bias in their relative abundance with depth, which we attribute to their organic test that is less prone to decomposition. Therefore, it is critical to include rare taxa within future transfer function models to incorporate their environmental tolerances and improve model predictions.

How does lead selection impact diagnosis through AI? Implications for continuous monitoring systems using convolutional neural networks

Abstract Background and Objectives Recently, there has been a surge in non-invasive diagnosing methods leveraging the dipolar 12-lead ECG information using Convolutional Neural Networks (CNNs). However, the traditional ECG contains 12 projections of the cardiac dipole in a three-dimensional space can result in redundancy. Furthermore, in continuous monitoring systems such as Holter and wearables, not all 12 leads are consistently available. This study aims to quantify redundancy and determine the best lead configuration for improved classification results using a CNN. It evaluates the level of redundancy in ECG leads and its impact on CNN classification. Methods The degree of redundancy was quantified by means of lead entropies (H) as depicted in Figure 1.A. The standard 12-lead ECG was chosen as the baseline input. From it, various subsets of leads were selected (6, 3, 1 leads) with the objective of attaining maximal orthogonality between leads while minimizing redundancy (Figure 1.B). A 2D-CNN model was trained for multiclass classification: Hypertension (HYP), ST-T interval Changes (STTC), Conduction Disturbances (CD), Myocardial Infarction (MI) and Normal (NORM). The designed inputs were employed to evaluate how redundancy impacts the performance of the CNN (see Figure 1.C). Results Reducing redundancy in input leads led to notable improvements in model prediction for all pathological conditions in terms of Area Under the Curve (AUC). Reducing to 6 and 8 leads were the most suitable input for the CNN (Table 1) increasing by 1.18% in HYP, 0.23% in STTC, 0.52% in CD, and 0.33% in MI. However, a significant reduction in redundancy may result in the loss of essential clinical information for classification, as observed in scenarios involving only 3 or 1 leads. Conclusions When diagnosing pathologies with systems featuring fewer leads, its combinations should aim for maximum orthogonality to reduce redundancy while preserving unique information. By strategically selecting a set of leads with reduced input redundancy, the effectiveness of the standard 12-lead ECG can be improved, thereby enhancing the utility of such continuous monitoring systems for automated diagnosis and also decreasing training and prediction times.

Enhancing Sustainability in Watershed Management: Spatiotemporal Assessment of Baseflow Alpha Factor in SWAT

The increasing frequency of extreme rainfall events poses significant challenges to sustainable water resource management, leading to severe natural disasters. To mitigate these challenges, understanding the hydrological characteristics of watersheds, especially baseflow, is critical for enhancing watershed resilience and supporting sustainable water quality and resource management. However, conventional watershed models often neglect the accurate simulation of baseflow recession. This study proposes a method for calculating and applying the alpha factor for each hydrologic response unit (HRU) in the Soil and Water Assessment Tool (SWAT), considering both temporal and spatial variability in baseflow. The study watershed has undergone significant development, increasing the need for effective water management strategies that promote long-term sustainability. The alpha factor was computed using BFlow2021, and its effectiveness was evaluated by comparing recession and baseflow estimates under different methods. The results indicate that incorporating monthly HRU-specific alpha factors significantly improves model predictions of recession characteristics, highlighting the need for a more spatially and temporally detailed approach in hydrological modeling. The proposed methodology can help clarify the connection between recession and baseflow and can be applied to ungauged stations, offering a valuable tool for sustainable watershed and water quality management.

Preoperative identification of early extrahepatic recurrence after hepatectomy for colorectal liver metastases: A machine learning approach.

Machine learning (ML) may provide novel insights into data patterns and improve model prediction accuracy. The current study sought to develop and validate an ML model to predict early extra-hepatic recurrence (EEHR) among patients undergoing resection of colorectal liver metastasis (CRLM). Patients with CRLM who underwent curative-intent resection between 2000 and 2020 were identified from an international multi-institutional database. An eXtreme gradient boosting (XGBoost) model was developed to estimate the risk of EEHR, defined as extrahepatic recurrence within 12months after hepatectomy, using clinicopathological factors. The relative importance of factors was determined using Shapley additive explanations (SHAP) values. Among 1410 patients undergoing curative-intent resection, 131 (9.3%) patients experienced EEHR. Median OS among patients with and without EEHR was 35.4months (interquartile range [IQR] 29.9-46.7) versus 120.5months (IQR 97.2-134.0), respectively (p<0.001). The ML predictive model had c-index values of 0.77 (95% CI, 0.72-0.81) and 0.77 (95% CI, 0.73-0.80) in the entire dataset and the validation data set with bootstrapping resamples, respectively. The SHAP algorithm demonstrated that T and N primary tumor categories, as well as tumor burden score were the three most important predictors of EEHR. An easy-to-use risk calculator for EEHR was developed and made available online at: https://junkawashima.shinyapps.io/EEHR/. An easy-to-use online calculator was developed using ML to help clinicians predict the chance of EEHR after curative-intent resection for CRLM. This tool may help clinicians in decision-making related to treatment strategies for patients with CRLM.

Interpretable and Physicochemical-Intuitive Deep Learning Approach for the Design of Thermal Resistance of Energetic Compounds.

Thermal resistance of energetic materials is critical due to its impact on safety and sustainability. However, developing predictive models remains challenging because of data scarcity and limited insights into quantitative structure-property relationships. In this work, a deep learning framework, named EM-thermo, was proposed to address these challenges. A data set comprising 5029 CHNO compounds, including 976 energetic compounds, was constructed to facilitate this study. EM-thermo employs molecular graphs and direct message-passing neural networks to capture structural features and predict thermal resistance. Using transfer learning, the model achieves an accuracy of approximately 97% for predicting the thermal-resistance property (decomposition temperatures above 573.15 K) in energetic compounds. The involvement of molecular descriptors improved model prediction. These findings suggest that EM-thermo is effective for correlating thermal resistance from the atom and covalent bond level, offering a promising tool for advancing molecular design and discovery in the field of energetic compounds.

Stance Detection in the Context of Fake News—A New Approach

Online social networks (OSNs) are inundated with an enormous daily influx of news shared by users worldwide. Information can originate from any OSN user and quickly spread, making the task of fact-checking news both time-consuming and resource-intensive. To address this challenge, researchers are exploring machine learning techniques to automate fake news detection. This paper specifically focuses on detecting the stance of content producers—whether they support or oppose the subject of the content. Our study aims to develop and evaluate advanced text-mining models that leverage pre-trained language models enhanced with meta features derived from headlines and article bodies. We sought to determine whether incorporating the cosine distance feature could improve model prediction accuracy. After analyzing and assessing several previous competition entries, we identified three key tasks for achieving high accuracy: (1) a multi-stage approach that integrates classical and neural network classifiers, (2) the extraction of additional text-based meta features from headline and article body columns, and (3) the utilization of recent pre-trained embeddings and transformer models.

Co-seismic landslide susceptibility mapping for the Luding earthquake area based on heterogeneous ensemble machine learning models

ABSTRACT Considering the complexity of landslide susceptibility mapping (LSM) at the municipal-scale, it is difficult for the assumption space of a single machine learning (ML) algorithm to fully and accurately reflect the intrinsic correlation between the landslide influencing factors and the actual catastrophic events under all scenarios. This study addresses this by integrating three heterogeneous ensemble learning (HEL) algorithms(Bagging, Blending, and Stacking) with four ML models (Gradient Boosting Decision Tree (GBDT), Extreme Gradient Boosting (XGBoost), K-Nearest Neighbors (KNN), and Backpropagation-Neural Network (BPNN)), to construct 27 HEL LSM models, and to realize optimization and selection of the models and algorithms through the comparative analysis of the results of the LSM and the accuracy of the prediction of the individual models. The final results indicate that the Bagging and Blending HEL algorithms fail to enhance model prediction accuracy effectively. In contrast, the Stacking HEL algorithm significantly improves model prediction accuracy (with an accuracy increase of over 1.4% and AUC values above 0.95). The Stacking (GBDT-XGBoost) model had the highest prediction accuracy (AUC = 0.964) and thus was selected as the optimal LSM model. This study's modeling and mapping results are intended to serve as a reference for geohazard prevention and sustainability studies in related cities.

Predicting the risk of diabetes complications using machine learning and social administrative data in a country with ethnic inequities in health: Aotearoa New Zealand

BackgroundIn the age of big data, linked social and administrative health data in combination with machine learning (ML) is being increasingly used to improve prediction in chronic disease, e.g., cardiovascular diseases (CVD). In this study we aimed to apply ML methods on extensive national-level health and social administrative datasets to assess the utility of these for predicting future diabetes complications, including by ethnicity.MethodsFive ML models were used to predict CVD events among all people with known diabetes in the population of New Zealand, utilizing nationwide individual-level administrative data.ResultsThe Xgboost ML model had the best predictive power for predicting CVD events three years into the future among the population with diabetes (N = 145,600). The optimization procedure also found limited improvement in prediction by ethnicity (using area under the receiver operating curve, [AUC]). The results indicated no trade-off between model predictive performance and equity gap of prediction by ethnicity (that is improving model prediction and reducing performance gaps by ethnicity can be achieved simultaneously). The list of variables of importance was different among different models/ethnic groups, for example: age, deprivation (neighborhood-level), having had a hospitalization event, and the number of years living with diabetes.Discussion and conclusionsWe provide further evidence that ML with administrative health data can be used for meaningful future prediction of health outcomes. As such, it could be utilized to inform health planning and healthcare resource allocation for diabetes management and the prevention of CVD events. Our results may suggest limited scope for developing prediction models by ethnic group and that the major ways to reduce inequitable health outcomes is probably via improved delivery of prevention and management to those groups with diabetes at highest need.

Sediment source tracing in a Brazilian subtropical catchment using diffuse reflectance: Effect of spectral ranges, pre-processing techniques, and multivariate model

ABSTRACT Agriculture intensification in Southern Brazil’s subtropical regions combined with the frequent occurrence of erosive rainfall has rendered the area a global water erosion hotspot. In this scenario, understanding and regulating erosion processes at the river catchment scale is critical for mitigating soil and water resource degradation. Traditional methods for tracing sediment sources are expensive and time-consuming and justify the development of alternative approaches. Therefore, in this study, we employed diffuse reflectance spectroscopy analyses in the ultraviolet-visible (UV-VIS), near-infrared (NIR), and mid-infrared (MIR) ranges, combined with multivariate models and spectral pre-processing techniques to estimate sediment source contributions in a homogeneous subtropical catchment (Conceição River, 804 km²). Soil samples (n = 181) were collected to characterize the four potential sediment sources, including: cropland (n = 78), stream bank (n = 36), unpaved road (n = 40) and pasture (n = 27). Moreover, 44 sediment samples were collected, including suspended sediment (n = 8), fine sediment deposited on the riverbed (n = 15), and suspended sediment samples collected in the water column during storm events (n = 21). Vector machine (SVM) model outperformed the others, with better accuracy and reliability. While UV-VIS spectra proved less effective due to soil homogeneity across the catchment, NIR and MIR spectra provided valuable information for discriminating sediment sources. Furthermore, reducing the number of potential sources (from four to three or two) improved model predictions, especially when distinguishing between surface sources (cropland and pasture) and subsurface sources (unpaved roads and stream banks). The study’s findings shed light on the power of efficient and cost-effective alternative methods for assessing sediment sources, which are vital for promoting effective erosion control and sustainable land management in similar regions.

Investigating colour vision and its structural correlates 15 years following a first demyelinating event

BackgroundWe investigated the long-term colour and contrast vision outcomes, 15 years after a first demyelinating event, with their structural correlates using optical coherence tomography (OCT) and brain MRI.MethodsPatients recruited with...

Comparison of engineered nanoparticle transport in columns of three different lengths: Transport experiments and multi-observation point modeling

This study focuses on the effectiveness of commonly-used 15 cm column lengths for investigating nanoparticle transport in porous media. Experimental tests examined the transport and retention behaviors of two types of nanoparticles, graphene oxide (GO) and titanium dioxide (TiO2) nanoparticles, in saturated sand columns of different lengths (15, 30 and 45 cm), while considering key environmental factors like ionic strength (IS, 1–50 mM), flow rate (1–3 mL min−1), and grain size (150–850 μm). In the 15 cm columns, both GO and TiO2 transport decreased with higher IS and lower flow rate; grain size affected GO and TiO2 differently. Smaller grain size increased GO retention in the sand columns through straining, thus weakening GO mobility, whereas increased fluid shear force suppressed the ripening of TiO2, enhancing its migration. Similar environmental effects were noted in longer columns (30 and 45 cm), but fitted transport parameters (Smax and k) and predicted long-term mobility (Lmax) indicated that 15 cm columns might underestimate nanoparticle mobility. Blocking and ripening models based on single and multiple observation points to simulate nanoparticle transport and retention showed that predictions aligned well with experimental data. These results indicate that using combinations of columns of different lengths to achieve multiple observation points improves model prediction accuracy; in single-column experiments, the 45 cm and 30 cm columns respectively better predict the mobility range of GO and TiO2 compared to 15 cm columns.

Genomic prediction for potato (Solanum tuberosum) quality traits improved through image analysis.

Potato (Solanum tuberosum L.) is the most widely grown vegetable in the world. Consumers and processors evaluate potatoes based on quality traits such as shape and skin color, making these traits important targets for breeders. Achieving and evaluating genetic gain is facilitated by precise and accurate trait measures. Historically, quality traits have been measured using visual rating scales, which are subject to human error and necessarily lump individuals with distinct characteristics into categories. Image analysis offers a method of generating quantitative measures of quality traits. In this study, we use TubAR, an image-analysis R package, to generate quantitative measures of shape and skin color traits for use in genomic prediction. We developed and compared different genomic models based on additive and additive plus non-additive relationship matrices for two aspects of skin color, redness, and lightness, and two aspects of shape, roundness, and length-to-width ratio for fresh market red and yellow potatoes grown in Minnesota between 2020 and 2022. Similarly, we used the much larger chipping potato population grown during the same time to develop a multi-trait selection index including roundness, specific gravity, and yield. Traits ranged in heritability with shape traits falling between 0.23 and 0.85, and color traits falling between 0.34 and 0.91. Genetic effects were primarily additive with color traits showing the strongest effect (0.47), while shape traits varied based on market class. Modeling non-additive effects did not significantly improve prediction models for quality traits. The combination of image analysis and genomic prediction presents a promising avenue for improving potato quality traits.

Improving prediction models of amyotrophic lateral sclerosis (ALS) using polygenic, pre-existing conditions, and survey-based risk scores in the UK Biobank.

Amyotrophic lateral sclerosis (ALS) causes profound impairments in neurological function, and a cure for this devastating disease remains elusive. This study aimed to identify pre-disposing genetic, phenotypic, and exposure-related factors for amyotrophic lateral sclerosis using multi-modal data and assess their joint predictive potential. Utilizing data from the UK (United Kingdom) Biobank, we analyzed an unrelated set of 292 ALS cases and 408,831 controls of European descent. Two polygenic risk scores (PRS) are constructed: "GWAS Hits PRS" and "PRS-CS," reflecting oligogenic and polygenic ALS risk profiles, respectively. Time-restricted phenome-wide association studies (PheWAS) were performed to identify pre-existing conditions increasing ALS risk, integrated into phenotypic risk scores (PheRS). A poly-exposure score ("PXS") captures the influence of environmental exposures measured through survey questionnaires. We evaluate the performance of these scores for predicting ALS incidence and stratifying risk, adjusting for baseline demographic covariates. Both PRSs modestly predicted ALS diagnosis but with increased predictive power when combined (covariate-adjusted receiver operating characteristic [AAUC] = 0.584 [0.525, 0.639]). PheRS incorporated diagnoses 1year before ALS onset (PheRS1) modestly discriminated cases from controls (AAUC = 0.515 [0.472, 0.564]). The "PXS" did not significantly predict ALS. However, a model incorporating PRSs and PheRS1 improved the prediction of ALS (AAUC = 0.604 [0.547, 0.667]), outperforming a model combining all risk scores. This combined risk score identified the top 10% of risk score distribution with a fourfold higher ALS risk (95% CI [2.04, 7.73]) versus those in the 40%-60% range. By leveraging UK Biobank data, our study uncovers pre-disposing ALS factors, highlighting the improved effectiveness of multi-factorial prediction models to identify individuals at highest risk for ALS.

A machine learning method based on TPE-XGBoost model for TRIP/TWIP near-β titanium alloy design

The traditional design method for TRIP/TWIP near-β titanium alloys is time-consuming and expensive. This paper proposes a machine-learning design method for a near-β titanium alloy that uses the Tree-structured Parzen Estimator (TPE) algorithm to optimize the hyperparameter space of XGBoost for improved model prediction accuracy in TRIP/TWIP. The model predicts the mechanical properties and plasticity mechanisms of the alloy based on its composition, key empirical parameters derived from the composition, and solution treatment temperature. The test results of the TPE-XGBoost model showed that when predicting the ultimate tensile strength (UTS), the adjusted determination coefficient (Adjusted R2), root mean square error (RMSE), and mean absolute error (MAE) values were 0.9264, 39.63 MPa and 44.83 MPa, respectively. When predicting elongation (El), the Adjusted R2, RMSE, and MAE values are 0.9134, 1.97 %, and 1.89 %, respectively. When predicting the plastic mechanisms of the alloy, the F1-score, Recall, Precision, Accuracy, and Area Under the ROC Curve (AUC) were 0.74, 0.75, 0.80, 0.78, and 0.83, respectively. A near-β titanium alloy, Ti-8V-4Mo-3.5Al-3Cr-2Zr (wt%), along with an associated heat treatment process, has been developed using this model. The experimental results have demonstrated that the model accurately predicts the strength, elongation, and plasticity mechanisms of the alloys. The TRIP/TWIP effects occur due to {332}<113> twinning and secondary α" martensite formation during tensile deformation. This study offers an alternative approach for designing TRIP/TWIP near-β titanium alloys.

An improved prediction model for miscibility characterization and optimization of CO2 and associated gas co-injection

Insufficient reservoir pressure and gas source problems seriously limit effective applications of CO2 flooding in hydrocarbon reservoirs. An attractive strategy is to co-inject the produced associated hydrocarbon gas with CO2 to achieve miscible/near-miscible flooding to enhance oil recovery. However, the composition of the injected mixture and the in-situ spatio-temporal composition evolution of the displacing/displaced fluids significantly affect the CO2-oil miscibility and the overall hydrocarbon recovery performance. In this work, based on slim tube simulation results, a prediction model of minimum miscible pressure (MMP)/minimum near-miscible pressure (MNMP) is established. This model takes into account the effects of CO2 concentration distribution, oil composition and temperature changes to achieve accurate spatiotemporal characterization of miscibility in the reservoir. In addition, the distribution law of the phase and components front for the impure CO2 injection process was also investigated. The miscible volume factor and recovery efficiency were used as the basis for judging the minimum injection concentration limit of CO2. The simulation results of slim tube experiment indicate that the MMP and MNMP obtained are 16.74 MPa and 12.35 MPa, respectively. Furthermore, the error caused by the traditional slim tube experiment can be reduced by using the CO2 produced fraction. It is found that CO2 injection concentration and reservoir temperature are the main factors affecting miscibility, but injection rate has less effect. Moreover, the fast migration of CH4 front leads to the shrinkage of miscible/near-miscible zone, which is not conducive to CO2 flooding. Finally, the CO2 injection concentration of non-breakthrough stage, breakthrough stage and large-scale breakthrough stage was determined to be 78.3%, 61.5% and 29.2%, respectively. This work provides strong evidence for the treatment of associated gas reinjection, which is beneficial for optimization of carbon capture, EOR-utilization and storage (CCUS-EOR) projects.

Broadband metamaterial linear polarization converter designed by a hybrid neural network with data augmentation

Deep learning techniques provide a new approach to the design and optimization of electromagnetic metamaterials. This study used a convolutional neural network and long short-term memory (CNN–LSTM) hybrid network to design and optimize a broadband metamaterial reflective linear polarization converter. The data augmentation method was also employed in few-shot learning to reduce optimization costs and improve model prediction performance. With the inverse prediction, a linear polarization converter that perfectly covers the Ku-band was obtained and fabricated with flexible printed circuit technology. Both simulation and experimental results indicate that this network can accurately predict the structural parameters. The polarization converter not only achieves remarkable broadband polarization conversion efficiency spanning the 2.2–18 GHz range but also maintains precise cross-polarization control across the entire Ku-band. The mean polarization conversion ratio in the Ku-band was calculated to be an impressive 99.69%. Finally, the mechanism of polarization conversion and the influence of each structural parameter on its performance further verify the optimality of the inverse design model. The use of CNN–LSTM deep learning methods significantly simplified the design process of electromagnetic metamaterials, reducing design costs while ensuring high design precision and excellent performance.

Assessing the Capabilities of UV-NIR Spectroscopy for Predicting Macronutrients in Hydroponic Solutions with Single-Task and Multi-Task Learning

Macronutrients, including nitrogen (N), phosphorus (P), potassium (K), calcium (Ca), magnesium (Mg), and sulfur (S), are the most basic nutrient elements in the solution for the hydroponic system. However, the current management of hydroponic nutrient solutions usually depends on EC and pH sensors due to the lack of accurate specific macronutrient sensing equipment, which easily leads to nutritional imbalance for the cultivated plant. In this study, the UV-NIR absorption spectroscopy (200–1100 nm) was used to predict six macronutrients in hydroponic solutions; two kinds of single-task learning algorithms, including partial least squares (PLS) and least absolute shrinkage and selection operator (LASSO), and two kinds of multi-task learning algorithms, including dirty multi-task learning (DMTL) and robust multi-task learning (RMTL), were investigated to develop prediction models and assess capabilities of UV-NIR. The results showed that N and Ca could be quantitatively predicted by UV-NIR with the ratio of performance to deviation (RPD) more than 2, K could be qualitatively predicted (1.4 &lt; RPD &lt; 2), and P, Mg, and S could not be successfully predicted (RPD &lt; 1.4); the RMTL algorithm outperformed others for predicting K and Ca benefit from the underlying task relationships with N; and predicting P, Mg, and S were identified as irrelevant (outlier) tasks. Our study provides a potential approach for predicting several macronutrients in hydroponic solutions with UV-NIR, especially using RMTL to improve model prediction ability.