Feature Importance Analysis Research Articles

Seismic Response Prediction of Rigid Rocking Structures Using Explainable LightGBM Models

This study emphasizes the explainability of machine learning (ML) models in predicting the seismic response of rigid rocking structures, specifically using the LightGBM algorithm. By employing SHapley Additive exPlanations (SHAP), partial dependence plots (PDP), and accumulated local effects (ALE), a comprehensive feature importance analysis has been performed. This revealed that ground motion parameters, particularly peak ground acceleration (PGA), are critical for predicting small rotations, while structural parameters like slenderness and frequency are more significant for larger rotations. Utilizing an extensive dataset generated from nonlinear time history analyses, the trained LightGBM model demonstrated high accuracy in estimating the maximum rotation angle of rigid blocks under natural ground motions. The study also examined the sensitivity of model performance to lower bound thresholds of the target variable, revealing that reduced feature sets can maintain predictive performance effectively. These findings advance ML-based modeling of seismic rocking responses, providing interpretable and accurate models that enhance our understanding of rocking structures’ dynamic behavior, which is crucial for designing resilient structures and improving seismic risk assessments. Future research will focus on incorporating additional parameters and exploring advanced ML techniques to further refine these models.

Designing of potential polymer donors with lower reorganization energy and chemical space exploration in search of materials for efficient organic solar cells

Machine learning with fascinating material predicting capabilities has emerged as a growing throughput artificial intelligence technology to accelerate the design, screening and prediction of material properties. In this study, chemical descriptors are employed to train machine learning (ML) models for the designing of potential polymer donor materials aimed at enhancing the efficiency of organic solar cells (OSCs). To achieve this, ten ML models are evaluated. Among all, the random forest regressor model has shown better predictive ability. In addition, to quantify the importance of features, the numerical scores of the significance of the features are determined through Feature importance analysis. Moreover, 10k new polymers are generated employing the python-based method and their reorganization energies are further predicted using a pre-trained ML model. The assemblage of polymers is done on the basis of the reorganization energy, those with lower reorganization energy being preferred. Furthermore, the synthetic accessibility and thus the structural diversity among the chosen polymers donors are examined, revealing a structural difference among the chosen polymers. This work paves a way for the designing and screening of efficient semiconductors with low reorganization energy to increase the efficiency of OSCs.

Uncovering corporate greenwashing: a predictive model based on Chinese heavy-pollution industries

Purpose As the global emphasis on environmental consciousness intensifies, many corporations claim to be environmentally responsible. However, some merely partake in “greenwashing” – a facade of eco-responsibility. Such deceptive behavior is especially prevalent in Chinese heavy-pollution industries. To counter these deceptive practices, this study aims to use machine learning (ML) techniques to develop predictive models against corporate greenwashing, thus facilitating the sustainable development of corporations. Design/methodology/approach This study develops effective predictive models for greenwashing by integrating multifaceted data sets, which include corporate external, organizational and managerial characteristics, and using a range of ML algorithms, namely, linear regression, random forest, K-nearest neighbors, support vector machines and artificial neural network. Findings The proposed predictive models register an improvement of over 20% in prediction accuracy compared to the benchmark value, furnishing stakeholders with a robust tool to challenge corporate greenwashing behaviors. Further analysis of feature importance, industry-specific predictions and real-world validation enhances the model’s interpretability and its practical applications across different domains. Practical implications This research introduces an innovative ML-based model designed to predict greenwashing activities within Chinese heavy-pollution sectors. It holds potential for application in other emerging economies, serving as a practical tool for both academics and practitioners. Social implications The findings offer insights for crafting informed, data-driven policies to curb greenwashing and promote corporate responsibility, transparency and sustainable development. Originality/value While prior research mainly concentrated on the factors influencing greenwashing behavior, this study takes a proactive approach. It aims to forecast the extent of corporate greenwashing by using a range of multi-dimensional variables, thus providing enhanced value to stakeholders. To the best of the authors’ knowledge, this is the first study introducing ML-based models designed to predict a company’s level of greenwashing.

Innovative graph neural network approach for predicting soil heavy metal pollution in the Pearl River Basin, China

Predicting soil heavy metal (HM) content is crucial for monitoring soil quality and ensuring ecological health. However, existing methods often neglect the spatial dependency of data. To address this gap, our study introduces a novel graph neural network (GNN) model, Multi-Scale Attention-based Graph Neural Network for Heavy Metal Prediction (MSA-GNN-HMP). The model integrates multi-scale graph convolutional network (MS-GCN) and attention-based GNN (AGNN) to capture spatial relationships. Using surface soil samples from the Pearl River Basin, we evaluate the MSA-GNN-HMP model against four other models. The experimental results show that the MSA-GNN-HMP model has the best predictive performance for Cd and Pb, with a coefficient of determination (R2) of 0.841 for Cd and 0.886 for Pb, and the lowest mean absolute error (MAE) of 0.403 mg kg−1 for Cd and 0.670 mg kg−1 for Pb, as well as the lowest root mean square error (RMSE) of 0.563 mg kg−1for Cd and 0.898 mg kg−1 for Pb. In feature importance analysis, latitude and longitude emerged as key factors influencing the heavy metal content. The spatial distribution prediction trend of heavy metal elements by different prediction methods is basically consistent, with the high-value areas of Cd and Pb respectively distributed in the northwest and northeast of the basin center. However, the MSA-GNN-HMP model demonstrates superior detail representation in spatial prediction. MSA-GNN-HMP model has excellent spatial information representation capabilities and can more accurately predict heavy metal content and spatial distribution, providing a new theoretical basis for monitoring, assessing, and managing soil pollution.

Deep learning and explainable artificial intelligence for investigating dental professionals' satisfaction with CAD software performance.

This study aimed to examine the satisfaction of dental professionals, including dental students, dentists, and dental technicians, with computer-aided design (CAD) software performance using deep learning (DL) and explainable artificial intelligence (XAI)-based behavioral analysis concepts. This study involved 436 dental professionals with diverse CAD experiences to assess their satisfaction with various dental CAD software programs. Through exploratory factor analysis, latent factors affecting user satisfaction were extracted from the observed variables. A multilayer perceptron artificial neural network (MLP-ANN) model was developed along with permutation feature importance analysis (PFIA) and the Shapley additive explanation (Shapley) method to gain XAI-based insights into individual factors' significance and contributions. The MLP-ANN model outperformed a standard logistic linear regression model, demonstrating high accuracy (95%), precision (84%), and recall rates (84%) in capturing complex psychological problems related to human attitudes. PFIA revealed that design adjustability was the most important factor impacting dental CAD software users' satisfaction. XAI analysis highlighted the positive impacts of features supporting the finish line and crown design, while the number of design steps and installation time had negative impacts. Notably, finish-line design-related features and the number of design steps emerged as the most significant factors. This study sheds light on the factors influencing dental professionals' decisions in using and selecting CAD software. This approach can serve as a proof-of-concept for applying DL-XAI-based behavioral analysis in dentistry and medicine, facilitating informed software selection and development.

Detecting depression severity using weighted random forest and oxidative stress biomarkers

This study employs machine learning to detect the severity of major depressive disorder (MDD) through binary and multiclass classifications. We compared models that used only biomarkers of oxidative stress with those that incorporate sociodemographic and health-related factors. Data collected from 830 participants, based on the Patient Health Questionnaire (PHQ-9) score, inform our analysis. In binary classification, the Random Forest (RF) classifier achieved the highest Area Under the Curve (AUC) of 0.84 when all features were included. In multiclass classification, the AUC improved from 0.84 with only oxidative stress biomarkers to 0.88 when all characteristics were included. To address data imbalance, weighted classifiers, and Synthetic Minority Over-sampling Technique (SMOTE) approaches were applied. Weighted random forest (WRF) improved multiclass classification, achieving an AUC of 0.91. Statistical tests, including the Friedman test and the Conover post-hoc test, confirmed significant differences between model performances, with WRF using all features outperforming others. Feature importance analysis shows that oxidative stress biomarkers, particularly GSH, are top ranked among all features. Clinicians can leverage the results of this study to improve their decision-making processes by incorporating oxidative stress biomarkers in addition to the standard criteria for depression diagnosis.

Predicting Emission Wavelengths in Benzobisoxazole-Based OLEDs with Gradient Boosted Ensemble Models.

We demonstrate the use of gradient-boosted ensemble models that accurately predict emission wavelengths in benzobis[1,2-d:4,5-d']oxazole (BBO) based fluorescent emitters. We have curated a database of 50 molecules from previously published data by the Jeffries-EL group using density functional theory (DFT) computed ground and excited state features. We consider two machine learning (ML) models based on (i) whole cruciform molecules and (ii) their constituent fragment molecules. Both ML models provide accurate predictions with root-mean-square errors between 30 and 36 nm, competitive with state-of-the-art deep learning models trained on orders of magnitude more molecules, and this accuracy holds even when tested on four new BBO emitters unseen by the models. We also provide an interpretable feature importance analysis and discuss the relevant relationships between DFT and changes in predicted emission wavelength.

Mapping global distributions of clay-size minerals via soil properties and machine learning techniques

Clay-size mineral is a vital ingredient of soil that influences various environment behaviors. It is crucial to establish a global distribution map of clay-size minerals to improve the recognition of environment variations. However, there is a huge gap of lacking some mineral contents in poorly accessible remote areas. In this work, machine learning (ML) approaches were conducted to predict the mineral contents and analyze their global abundance changes through the relationship between soil properties and mineral distributions. The average content of kaolinite, illite, smectite, vermiculite, chlorite, and feldspar were predicated to be 28.69 %, 22.30 %, 12.42 %, 5.43 %, 5.03 %, and 1.44 % respectively. Model interpretation showed that topsoil bulk density and drainage class were the most significant factors for predicting all six minerals. It could be seen from the feature importance analysis that bulk density notably reflected the distribution of 2:1 layered minerals more than that of 1:1 mineral. High drainage favored secondary minerals development, while low drainage was more benefited for primary minerals. Moreover, the content variation of different minerals aligned with the distribution of corresponding soil properties, which affirmed the accuracy of established models. This study proposed a new approach to predict mineral contents through soil properties, which filled a necessary step of understanding the geochemical cycles of soil-related processes.

Spatial instability of crash prediction models: A case of scooter crashes

Scooters have gained widespread popularity in recent years due to their accessibility and affordability, but safety concerns persist due to the vulnerability of riders. Researchers are actively investigating the safety implications associated with scooters, given their relatively new status as transportation options. However, analyzing scooter safety presents a unique challenge due to the complexity of determining safe riding environments. This study presents a comprehensive analysis of scooter crash risk within various buffer zones, utilizing the Extreme Gradient Boosting (XGBoost) machine learning algorithm. The core objective was to unravel the multifaceted factors influencing scooter crashes and assess the predictive model’s performance across different buffers or spatial proximity to crash sites. After evaluating the model’s accuracy, sensitivity, and specificity across buffer distances ranging from 5 ft to 250 ft with the scooter crash as a reference point, a discernible trend emerged: as the buffer distance decreases, the model’s sensitivity increases, although at the expense of accuracy and specificity, which exhibit a gradual decline. Notably, at the widest buffer of 250 ft, the model achieved a high accuracy of 97% and specificity of 99%, but with a lower sensitivity of 31%. Contrastingly, at the closest buffer of 5 ft, sensitivity peaked at 95%, albeit with slightly reduced accuracy and specificity. Feature importance analysis highlighted the most significant predictor across all buffer distances, emphasizing the impact of vehicle interactions on scooter crash likelihood. Explainable Artificial Intelligence through SHAP value analysis provided deeper insights into each feature’s contribution to the predictive model, revealing passenger vehicle types of significantly escalated crash risks. Intriguingly, specific vehicular maneuvers, notably stopping in traffic lanes, alongside the absence of Traffic Control Devices (TCDs), were identified as the major contributors to increased crash occurrences. Road conditions, particularly wet and dry, also emerged as substantial risk factors. Furthermore, the study highlights the significance of road design, where elements like junction types and horizontal alignments – specifically 4 and 5-legged intersections and curves – are closely associated with heightened crash risks. These findings articulate a complex and spatially detailed framework of factors impacting scooter crashes, offering vital insights for urban planning and policymaking.

Active learning streamlines development of high performance catalysts for higher alcohol synthesis

Developing efficient catalysts for syngas-based higher alcohol synthesis (HAS) remains a formidable research challenge. The chain growth and CO insertion requirements demand multicomponent materials, whose complex reaction dynamics and extensive chemical space defy catalyst design norms. We present an alternative strategy by integrating active learning into experimental workflows, exemplified via the FeCoCuZr catalyst family. Our data-aided framework streamlines navigation of the extensive composition and reaction condition space in 86 experiments, offering >90% reduction in environmental footprint and costs over traditional programs. It identifies the Fe65Co19Cu5Zr11 catalyst with optimized reaction conditions to attain higher alcohol productivities of 1.1 gHA h−1 gcat−1 under stable operation for 150 h on stream, a 5-fold improvement over typically reported yields. Characterization reveals catalytic properties linked to superior activities despite moderate higher alcohol selectivities. To better reflect catalyst demands, we devise multi-objective optimization to maximize higher alcohol productivity while minimizing undesired CO2 and CH4 selectivities. An intrinsic trade-off between these metrics is uncovered, identifying Pareto-optimal catalysts not readily discernible by human experts. Finally, based on feature-importance analysis, we formulate data-informed guidelines to develop performance-specific FeCoCuZr systems. This approach goes beyond existing HAS catalyst design strategies, is adaptable to broader catalytic transformations, and fosters laboratory sustainability.

Quantifying Predictive Uncertainty and Feature Selection in River Bed Load Estimation: A Multi-Model Machine Learning Approach with Particle Swarm Optimization

This study presents a comprehensive multi-model machine learning (ML) approach to predict river bed load, addressing the challenge of quantifying predictive uncertainty in fluvial geomorphology. Six ML models—random forest (RF), categorical boosting (CAT), extra tree regression (ETR), gradient boosting machine (GBM), Bayesian regression model (BRM), and K-nearest neighbors (KNNs)—were thoroughly evaluated across several performance metrics like root mean square error (RMSE), and correlation coefficient (R). To enhance model training and optimize performance, particle swarm optimization (PSO) was employed for hyperparameter tuning across all the models, leveraging its capability to efficiently explore complex hyperparameter spaces. Our findings indicated that RF, GBM, CAT, and ETR demonstrate superior predictive performance (R score > 0.936), benefiting significantly from PSO. In contrast, BRM displayed lower performance (0.838), indicating challenges with Bayesian approaches. The feature importance analysis, including permutation feature and SHAP values, highlighted the non-linear interdependencies between the variables, with river discharge (Q), bed slope (S), and flow width (W) being the most influential. This study also examined the specific impact of individual variables on model performance by adding and excluding individual variables, which is particularly meaningful when choosing input variables for the model, especially in limited data conditions. Uncertainty quantification through Monte Carlo simulations highlighted the enhanced predictability and reliability of models with larger datasets. The correlation between increased training data and improved model precision was evident in the consistent rise in mean R scores and reduction in standard deviations as the sample size increased. This research underscored the potential of advanced ensemble methods and PSO to mitigate the limitations of single-predictor models and exploit collective model strengths, thereby improving the reliability of predictions in river bed load estimation. The insights from this study provide a valuable framework for future research directions focused on optimizing ensemble configurations for hydro-dynamic modeling.

Interpretable Machine Learning Models for Prediction of UHPC Creep Behavior

The creep behavior of Ultra-High-Performance Concrete (UHPC) was investigated by machine learning (ML) and SHapley Additive exPlanations (SHAP). Important features were selected by feature importance analysis, including water-to-binder ratio, aggregate-to-cement ratio, compressive strength at loading age, elastic modulus at loading age, loading duration, steel fiber volume content, and curing temperature. Four typical ML models—Random Forest (RF), Artificial Neural Network (ANN), Extreme Gradient Boosting Machine (XGBoost), and Light Gradient Boosting Machine (LGBM)—were studied to predict the creep behavior of UHPC. Via Bayesian optimization and 5-fold cross-validation, the ML models were tuned to achieve high accuracy (R2 = 0.9847, 0.9627, 0.9898, and 0.9933 for RF, ANN, XGBoost, and LGBM, respectively). The contribution of different features to the creep behavior was ranked. Additionally, SHAP was utilized to interpret the predictions by the ML models, and four parameters stood out as the most influential for the creep coefficient: loading duration, curing temperature, compressive strength at loading age, and water-to-binder ratio. The SHAP results were consistent with theoretical understanding. Finally, the UHPC creep curves for three different cases were plotted based on the ML model developed, and the prediction by the ML model was more accurate than that by fib Model Code 2010.

Preparation of Thermochromic Vanadium Dioxide Films Assisted by Machine Learning.

In recent years, smart windows have attracted widespread attention due to their ability to respond to external stimuli such as light, heat, and electricity, thereby intelligently adjusting the ultraviolet, visible, and near-infrared light in solar radiation. VO2(M) undergoes a reversible phase transition from an insulating phase (monoclinic, M) to a metallic phase (rutile, R) at a critical temperature of 68 °C, resulting in a significant difference in near-infrared transmittance, which is particularly suitable for use in energy-saving smart windows. However, due to the multiple valence states of vanadium ions and the multiphase characteristics of VO2, there are still challenges in preparing pure-phase VO2(M). Machine learning (ML) can learn and generate models capable of predicting unknown data from vast datasets, thereby avoiding the wastage of experimental resources and reducing time costs associated with material preparation optimization. Hence, in this paper, four ML algorithms, namely multi-layer perceptron (MLP), random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGB), were employed to explore the parameters for the successful preparation of VO2(M) films via magnetron sputtering. A comprehensive performance evaluation was conducted on these four models. The results indicated that XGB was the top-performing model, achieving a prediction accuracy of up to 88.52%. A feature importance analysis using the SHAP method revealed that substrate temperature had an essential impact on the preparation of VO2(M). Furthermore, characteristic parameters such as sputtering power, substrate temperature, and substrate type were optimized to obtain pure-phase VO2(M) films. Finally, it was experimentally verified that VO2(M) films can be successfully prepared using optimized parameters. These findings suggest that ML-assisted material preparation is highly feasible, substantially reducing resource wastage resulting from experimental trial and error, thereby promoting research on material preparation optimization.

AI models predicting breast cancer distant metastasis using LightGBM with clinical blood markers and ultrasound maximum diameter

Breast cancer metastasis significantly impacts women's health globally. This study aimed to construct predictive models using clinical blood markers and ultrasound data to predict distant metastasis in breast cancer patients, ensuring clinical applicability, cost-effectiveness, relative non-invasiveness, and accessibility of these models. Analysis was conducted on data from 416 patients across two centers, focusing on clinical blood markers (tumor markers, liver and kidney function indicators, blood lipid markers, cardiovascular biomarkers) and maximum lesion diameter from ultrasound. Feature reduction was performed using Spearman correlation and LASSO regression. Two models were built using LightGBM: a clinical model (using clinical blood markers) and a combined model (incorporating clinical blood markers and ultrasound features), validated in training, internal test, and external validation (test1) cohorts. Feature importance analysis was conducted for both models, followed by univariate and multivariate regression analyses of these features. The AUC values of the clinical model in the training, internal test, and external validation (test1) cohorts were 0.950, 0.795, and 0.883, respectively. The combined model showed AUC values of 0.955, 0.835, and 0.918 in the training, internal test, and external validation (test1) cohorts, respectively. Clinical utility curve analysis indicated the combined model's superior net benefit in identifying breast cancer with distant metastasis across all cohorts. This suggests the combined model's superior discriminatory ability and strong generalization performance. Creatine kinase isoenzyme (CK-MB), CEA, CA153, albumin, creatine kinase, and maximum lesion diameter from ultrasound played significant roles in model prediction. CA153, CK-MB, lipoprotein (a), and maximum lesion diameter from ultrasound positively correlated with breast cancer distant metastasis, while indirect bilirubin and magnesium ions showed negative correlations. This study successfully utilized clinical blood markers and ultrasound data to develop AI models for predicting distant metastasis in breast cancer. The combined model, incorporating clinical blood markers and ultrasound features, exhibited higher accuracy, suggesting its potential clinical utility in predicting and identifying breast cancer distant metastasis. These findings highlight the potential prospects of developing cost-effective and accessible predictive tools in clinical oncology.

Buckling behaviors prediction of biological staggered composites with finite element analysis and machine learning coupled method

Staggered structure, where mineral crystals are arranged in a staggered manner in protein matrix, is the most representative microstructure in biological composites. Because of the elongated geometry and large aspect ratio of mineral crystals, their failure mechanism under compressive loading is predominantly controlled by buckling. The microstructural complexity, diversity of the constituent materials, and the complexity of buckling behaviors present substantial challenges for the precise forecasting of the buckling behavior of biological materials. In this paper, a novel method combining finite element analysis with machine learning is proposed. Specifically, based on the large dataset obtained using the finite element method, the buckling strength of biological staggered composites was rapidly predicted using a machine learning algorithm. Herein, the effects of microstructure and component materials on buckling behavior have been investigated systematically. The results indicate that the neural network models are highly accurate in predicting critical buckling strength. Aspect ratio and modulus ratio significantly affect the critical buckling strength of staggered structures according to the results of feature importance analysis. Additionally, the post-buckling modes of staggered structures exhibit heightened sensitivity to initial geometric defects, particularly in higher-order buckling modes. Our research provides valuable insights into fast and accurate prediction of the buckling behavior of biological composites and their design.

Rapid total sialic acid monitoring during cell culture process using a machine learning model based soft sensor.

Total sialic acid content (TSA) in biotherapeutic proteins is often a critical quality attribute as it impacts the drug efficacy. Traditional wet chemical assays to quantify TSA in biotherapeutic proteins during cell culture typically takes several hours or longer due to the complexity of the assay which involves isolation of sialic acid from the protein of interest, followed by sample preparation and chromatographic based separation for analysis. Here, we developed a machine learning model-based technology to rapidly predict TSA during cell culture by using typically measured process parameters. The technology features a user interface, where the users only have to upload cell culture process parameters as input variables and TSA values are instantly displayed on a dashboard platform based on the model predictions. In this study, multiple machine learning algorithms were assessed on our dataset, with the Random Forest model being identified as the most promising model. Feature importance analysis from the Random Forest model revealed that attributes like viable cell density (VCD), glutamate, ammonium, phosphate, and basal medium type are critical for predictions. Notably, while the model demonstrated strong predictability by Day 14 of observation, challenges remain in forecasting TSA values at the edges of the calibration range. This research not only emphasizes the transformative power of machine learning and soft sensors in bioprocessing but also introduces a rapid and efficient tool for sialic acid prediction, signaling significant advancements in bioprocessing. Future endeavors may focus on data augmentation to further enhance model precision and exploration of process control capabilities.

Can Economic, Geopolitical and Energy Uncertainty Indices Predict Bitcoin Energy Consumption? New Evidence from a Machine Learning Approach

This paper explores the predictive power of economic and energy policy uncertainty indices and geopolitical risks for bitcoin’s energy consumption. Three machine learning tools, SVR (scikit-learn 1.5.0),CatBoost 1.2.5 and XGboost 2.1.0, are used to evaluate the complex relationship between uncertainty indices and bitcoin’s energy consumption. Results reveal that the XGboost model outperforms both SVR and CatBoost in terms of accuracy and convergence. Furthermore, the feature importance analysis performed by the Shapley additive explanation (SHAP) method indicates that all uncertainty indices exhibit a significant capacity to predict bitcoin’s future energy consumption. Moreover, SHAP values suggest that economic policy uncertainty captures valuable predictive information from the energy uncertainty indices and geopolitical risks that affect bitcoin’s energy consumption.

Multi-objective explainable smart dispatch for integrated energy system based on an explainable MO-RL method

With the rapid development of renewable energy penetration and the transition to high-efficiency, low-carbon-footprint energy systems, renewable-based heat-electricity integrated energy systems (RHE-IES) have become a viable method. The dispatch and operation scheduling of the RHE-IES is more complex than before because a well-operated RHE-IES should be economically favorable, environmentally friendly, and feasible with respect to security constraints. However, it is not easy to balance these trade-offs with uncertainties. In this work, a multi-objective hierarchical deep reinforcement learning (H-DRL) method is proposed to address the dispatch of RHE-IES, and it is also explainable due to the design of the algorithmic structure, the decoupling of rewards, and the analysis of feature importance. First, the dispatch and operation scheduling problem of the RHE-IES is formulated considering economic, environmental, and security constraints and is converted to a Markov decision process. Second, a novel H-DRL method is proposed with a multi-critic, single-actor structure, where each objective is handled by a designated deep critic network with decoupled rewards, and the action value functions of different objectives are hierarchically optimized by the actor network. H-DRL can be trained in a comprehensive environment offline and provide fast online decisions considering the uncertainties of renewable generation. Third, a shapley additive explanations method is carried out utilizing the explainable nature of the H-DRL structure and the loss function design. The simulation results and case studies show the effectiveness of the proposed method in terms of balancing the tradeoffs of different objectives, overcoming uncertainties and enhancing AI interpretability.

Primary Controlling Factors of Apatite Trace Element Composition and Implications for Exploration in Orogenic Gold Deposits

AbstractSignificant and readily accessible orogenic gold deposits have been previously recognized, exploited, and progressively depleted. Innovative approaches are required to discover new and deeply buried deposits. Recently, trace element variations in apatite have been used to distinguish fertile and barren environments as reliable mineral exploration tools. In this study, machine learning models using Random Forest and Deep Neutral Network are utilized to assess the fertility of quartz veins and altered zones in the orogenic gold systems. The two models have been trained using trace element data of apatite, and the performance of both models yield good classification accuracy (∼90% on average) with low false positive rates. Feature importance analysis (Gini decrease and hidden layer weights) suggest that Pb, As, U, Sr, Eu, Mn, and Fe are the important parameters. Arsenic, U, Eu, Mn, and Fe are redox‐sensitive elements, with their concentrations responding to changes in fluid redox conditions. Strontium primarily originates from the breakdown of plagioclase, which is more likely to occur under oxidizing fluid conditions. Therefore, we infer that the main controlling factor of the models is the redox conditions. The two distinct models consistently highlight the most significant contribution of Pb to this differentiation, even though Pb is not a redox‐sensitive element and can only substitute for Ca2+ in apatite as Pb2+. We infer that the high contribution of Pb may be attributed to the potential transportation of Au in the form of a Pb‐(Bi)‐Au melt, and the Pb content in apatite is influenced by the Pb content in the melt, fluid oxygen, and sulfur fugacity. We also propose a novel discriminant plot using Linear Discriminant Analysis (LDA) to assess the mineralization potential in quartz veins and alteration zones based on apatite trace element data. The machine learning and LDA results suggest that apatite trace elements could be used effectively in the future orogenic gold deposit exploration.

Spectral enhancement of PlanetScope using Sentinal-2 images to estimate soybean yield and seed composition

Soybean is an essential crop to fight global food insecurity and is of great economic importance around the world. Along with genetic improvements aimed at boosting yield, soybean seed composition also changed. Since conditions during crop growth and development influences nutrient accumulation in soybean seeds, remote sensing offers a unique opportunity to estimate seed traits from the standing crops. Capturing phenological developments that influence seed composition requires frequent satellite observations at higher spatial and spectral resolutions. This study introduces a novel spectral fusion technique called multiheaded kernel-based spectral fusion (MKSF) that combines the higher spatial resolution of PlanetScope (PS) and spectral bands from Sentinel 2 (S2) satellites. The study also focuses on using the additional spectral bands and different statistical machine learning models to estimate seed traits, e.g., protein, oil, sucrose, starch, ash, fiber, and yield. The MKSF was trained using PS and S2 image pairs from different growth stages and predicted the potential VNIR1 (705 nm), VNIR2 (740 nm), VNIR3 (783 nm), SWIR1 (1610 nm), and SWIR2 (2190 nm) bands from the PS images. Our results indicate that VNIR3 prediction performance was the highest followed by VNIR2, VNIR1, SWIR1, and SWIR2. Among the seed traits, sucrose yielded the highest predictive performance with RFR model. Finally, the feature importance analysis revealed the importance of MKSF-generated vegetation indices from fused images.