Innovative data-driven approaches for environmental sciences

Permanent storage of anthropogenic CO2 in deep geologic formations is being considered as a means to reduce the concentration of atmospheric CO2 and thus its contribution to global climate change. To ensure safe and effective geologic sequestration, numerous studies have been completed of the extent to which the CO2 migrates within geologic formations and what physical and geochemical changes occur in these formations when CO2 is injected. Sophisticated, computerized reservoir simulations are used as part of field site and laboratory CO2 sequestration studies. These simulations use coupled multiphase flow-reactive chemical transport models and/or standalone (i.e., no coupled fluid transport) geochemical models to calculate gas solubility, aqueous complexation, reduction/oxidation (redox), and/or mineral solubility reactions related to CO2 injection and sequestration. Thermodynamic data are critical inputs to modeling geochemical processes. The adequacy of thermodynamic data for carbonate compounds has been identified as an important data requirement for the successful application of these geochemical reaction models to CO2 sequestration. A review of thermodynamic data for CO2 gas and carbonate aqueous species and minerals present in published data compilations and databases used in geochemical reaction models was therefore completed. Published studies that describe mineralogical analyses from CO2 sequestration field and natural analogue sites and laboratory studies were also reviewed to identify specific carbonate minerals that are important to CO2 sequestration reactions and therefore require thermodynamic data. The results of the literature review indicated that an extensive thermodynamic database exists for CO2 and CH4 gases, carbonate aqueous species, and carbonate minerals. Values of ΔfG298° and/or log Kr,298° are available for essentially all of these compounds. However, log Kr,T° or heat capacity values at temperatures above 298 K exist for less than approximately one-third of these compounds. Because the temperatures of host formations that will be used for CO2 injection and sequestration will be at tempera¬tures in the range of 50ºC to 100ºC or greater, the lack of high temperature thermodynamic values for key carbonate compounds especially minerals, will impact the accuracy of some modeling calculations.

Given the increasing importance of machine learning (ML) in our lives, several algorithmic fairness techniques have been proposed to mitigate biases in the outcomes of the ML models. However, most of these techniques are specialized to cater to a single family of ML models and a specific definition of fairness, limiting their adaptibility in practice. We introduce a general constrained Bayesian optimization (BO) framework to optimize the performance of any ML model while enforcing one or multiple fairness constraints. BO is a model-agnostic optimization method that has been successfully applied to automatically tune the hyperparameters of ML models. We apply BO with fairness constraints to a range of popular models, including random forests, gradient boosting, and neural networks, showing that we can obtain accurate and fair solutions by acting solely on the hyperparameters. We also show empirically that our approach is competitive with specialized techniques that enforce model-specific fairness constraints, and outperforms preprocessing methods that learn fair representations of the input data. Moreover, our method can be used in synergy with such specialized fairness techniques to tune their hyperparameters. Finally, we study the relationship between fairness and the hyperparameters selected by BO. We observe a correlation between regularization and unbiased models, explaining why acting on the hyperparameters leads to ML models that generalize well and are fair.

Over the decades, as industrialization progressed, energy has been a critical topic for scientists and engineers. Particularly, photovoltaic technology has drawn great attention in the renewable energy industry as an environmentally clean technology for converting sunlight into electricity. However, the complexity of energy chemistry and the need for novel materials to improve solar cell efficiency and cost-effectiveness have led to challenges in establishing rules beyond empirical observations. Machine learning models are being developed to streamline the prediction process and efficiently predict photovoltaic parameters. This paper proposes a novel hybrid-optimized multi-objective predictive model to predict the photovoltaic parameters: open-circuit voltage (Voc), current density (Jsc), fill factor (FF), and power conversion efficiency (PCE). The proposed model is based on Bayesian Optimization (BO) with the ensemble Bootstrap Aggregating (Bagging) decision tree. The proposed model integrates with the Explainable Artificial Intelligence (XAI) using the SHAP (Shapley Additive Explanations) values to introduce feature importance analysis that provides valuable insights into the impact of individual features on prediction outputs. The proposed model, named BO-Bagging, achieves high prediction accuracy, with an average high correlation coefficient of r = 0.92, a coefficient of determination of R2 = 0.82, and a Mean Square Error (MSE) of 0.00172. In terms of complexity, the BO-Bagging model has a short processing time that is indicated with an average training time of 182.7 s and an average inference time averaging 0.00062 s. Also, the number of predicted observations per second is measured by prediction speed, which results in good prediction accuracy with an average of 2188.4 and model size with an average of 10,740.4 KB. Finally, the proposed model’s primary critical operations across each phase, from training to predicting the final outputs, are represented by 108 floating-point operations per second (FLOPS). All of these results demonstrate the proposed model’s accuracy and high efficiency in intelligent chemical applications.

This study presents a data-driven approach for optimizing the operating altitude of Airborne Wind Energy (AWE) systems to maximize net energy production. Determining the optimal operating altitude of an AWE system is challenging, as the wind speed constantly varies with both time and altitude. Furthermore, without expensive auxiliary equipment, the wind speed is only measurable at the AWE system's operating altitude. The work presented in this paper shows how tools from machine learning can be blended with real-time control to optimize the AWE system's operating altitude efficiently, without the use of auxiliary wind profiling equipment. Specifically, Bayesian Optimization, which is a data-driven technique for finding the optimum of an unknown and expensive-to-evaluate objective function, is applied to the real-time control of an AWE system. The underlying objective function is modeled by a Gaussian Process (GP); then, Bayesian Optimization utilizes the predictive uncertainty information from the GP to decide the best subsequent operating altitude. In the AWE application, conventional Bayesian Optimization is extended to handle the time-varying nature of the wind shear profile (wind speed vs. time). Using real wind data, our method is validated against three baseline approaches. Our simulation results show that the Bayesian Optimization method is successful in dramatically increasing power production over these baselines.

Unfolding Explainable AI for Brain Tumor Segmentation

Explainable artificial intelligence received great research attention in the past few years during the widespread of Black-Box techniques in sensitive fields such as medical care, self-driving cars, etc. Artificial intelligence needs explainable methods to discover model biases. Explainable artificial intelligence will lead to obtaining fairness and Transparency in the model. Making artificial intelligence models explainable and interpretable is challenging when implementing black-box models. Because of the inherent limitations of collecting data in its raw form, data fusion has become a popular method for dealing with such data and acquiring more trustworthy, helpful, and precise insights. Compared to other, more traditional-based data fusion methods, machine learning's capacity to automatically learn from experience with nonexplicit programming significantly improves fusion's computational and predictive power. This paper comprehensively studies the most explainable artificial intelligent methods based on anomaly detection. We proposed the required criteria of the transparency model to measure the data fusion analytics techniques. Also, define the different used evaluation metrics in explainable artificial intelligence. We provide some applications for explainable artificial intelligence. We provide a case study of anomaly detection with the fusion of machine learning. Finally, we discuss the key challenges and future directions in explainable artificial intelligence.

This research investigates the use of explainable artificial intelligence (XAI) in ancient architecture and lacquer art. The aim is to create accurate and interpretable models to reveal these cultural artefacts’ underlying design principles and techniques. To achieve this, machine learning and data-driven techniques are employed, which provide new insights into their construction and preservation. The study emphasises the importance of transparent and trustworthy AI systems, which can enhance the reliability and credibility of the results. The developed model outperforms CNN-based emotion recognition and random forest models in all four evaluation metrics, achieving an impressive accuracy of 92%. This research demonstrates the potential of XAI to support the study and conservation of ancient architecture and lacquer art, opening up new avenues for interdisciplinary research and collaboration.

Metabolic dysfunction-associated steatotic liver disease (MASLD) is a leading cause of chronic liver disease globally. Current diagnostic methods, such as liver biopsies, are invasive and have limitations, highlighting the need for non-invasive alternatives. To investigate extracellular vesicles (EVs) as potential biomarkers for diagnosing and staging steatosis in patients with MASLD using machine learning (ML) and explainable artificial intelligence (XAI). In this single-center observational study, 798 patients with metabolic dysfunction were enrolled. Of these, 194 met the eligibility criteria, and 76 successfully completed all study procedures. Transient elastography was used for steatosis and fibrosis staging, and circulating plasma EV characteristics were analyzed through nanoparticle tracking. Twenty ML models were developed: Six to differentiate non-steatosis (S0) from steatosis (S1-S3); and fourteen to identify severe steatosis (S3). Models utilized EV features (size and concentration), clinical (advanced fibrosis and presence of type 2 diabetes mellitus), and anthropomorphic (sex, age, height, weight, body mass index) data. Their performance was assessed using receiver operating characteristic (ROC)-area under the curve (AUC), specificity, and sensitivity, while correlation and XAI analysis were also conducted. The CatBoost C1a model achieved an ROC-AUC of 0.71/0.86 (train/test) on average across ten random five-fold cross-validations, using EV features alone to distinguish S0 from S1-S3. The CatBoost C2h-21 model achieved an ROC-AUC of 0.81/1.00 (train/test) on average across ten random three-fold cross-validations, using engineered features including EVs, clinical features like diabetes and advanced fibrosis, and anthropomorphic data like body mass index and weight for identifying severe steatosis (S3). Key predictors included EV mean size and concentration. Correlation, XAI, and SHapley Additive exPlanations analysis revealed non-linear feature relationships with steatosis stages. The EV-based ML models demonstrated that the mean size and concentration of circulating plasma EVs constituted key predictors for distinguishing the absence of significant steatosis (S0) in patients with metabolic dysfunction, while the combination of EV, clinical, and anthropomorphic features improved the diagnostic accuracy for the identification of severe steatosis. The algorithmic approach using ML and XAI captured non-linear patterns between disease features and provided interpretable MASLD staging insights. However, further large multicenter studies, comparisons, and validation with histopathology and advanced imaging methods are needed.

Machine learning (ML) is a powerful tool for hydrological modelling, prediction, dataset creation and the generation of insights into hydrological processes. As such, ML has become integral to the field of large-sample hydrology, where hundreds to thousands of river catchments are included within a single ML model to capture diverse hydrological behaviours and improve model generalizability. This manuscript outlines recent advances in ML for large-sample hydrology. We review new tools in explainable AI (XAI) and interpretability approaches, as well as challenges in these areas. Key research avenues for large-sample hydrology include addressing variability in interpretations resulting from different ML models and XAI techniques, enhancing hydrological predictions in data-sparse and human-impacted regions, reducing the 'cascade of uncertainty' inherent in hydrological modelling, developing improved methods for multivariate prediction and identifying causal relationships.This article is part of the discussion meeting issue 'Hydrology in the 21st century: challenges in science, to policy and practice'.

Wireless cellular networks have many parameters that are normally tuned upon deployment and re-tuned as the network changes. Many operational parameters affect reference signal received power (RSRP), reference signal received quality (RSRQ), signal-to-interference-plus-noise-ratio (SINR), and, ultimately, throughput. In this paper, we develop and compare two approaches for maximizing coverage and minimizing interference by jointly optimizing the transmit power and downtilt (elevation tilt) settings across sectors. To evaluate different parameter configurations offline, we construct a realistic simulation model that captures geographic correlations. Using this model, we evaluate two optimization methods: deep deterministic policy gradient (DDPG), a reinforcement learning (RL) algorithm, and multi-objective Bayesian optimization (BO). Our simulations show that both approaches significantly outperform random search and converge to comparable Pareto frontiers, but that BO converges with two orders of magnitude fewer evaluations than DDPG. Our results suggest that data-driven techniques can effectively self-optimize coverage and capacity in cellular networks.

In recent decades, various software effort estimation (SEE) algorithms have been suggested. Unfortunately, generating high-precision accuracy is still a major challenge in the context of SEE. The use of traditional techniques and parametric approaches is largely inaccurate because they produce biased and subjective accuracy. Meanwhile, none of the machine learning methods performed well. This study applies the AdaBoost ensemble learning method and random forest (RF), on the other hand the Bayesian optimization method is applied to determine the hyperparameters of this model. The PROMISE repository and the ISBSG dataset were used to build the SEE model. The developed model was comprehensively compared with four machine learning methods (classification and regression tree, k-nearest neighbor, multilayer perceptron, and support vector regression) under 3-fold cross validation (CV). It can be seen that the RF method based on AdaBoost ensemble learning and bayesian optimization outperforms this approach. In addition, the AdaBoost-based model assigns a feature importance rating, which makes it a promising tool in software effort prediction.

<p>The modern advances in computing and experimental capabilities in the research of water-rock-interactions require geoscientists to routinely combine laboratory data and models to produce knowledge in order to solve pressing societal challenges connected to subsurface utilization. Data science is hence a more and more pervasive instrument also for  geochemists, which in turn demands flexible and easy to learn software adaptable to their specific needs. <br>In this contribution we showcase geochemical and reactive transport modelling with our RedModRphree [1] extension package for the GNU R environment and programming language. The new version of the package leverages the R interface to the established PHREEQC geochemical simulator maintained by its original authors [2]. R has established itself as de facto standard language for statistics and machine learning. It enjoys increasing diffusion in many applied scientific fields such as bioinformatics, chemometrics and ecological modelling. The availability of excellent third party extensions such as the thermodynamic package CHNOSZ [3], which extends the functionalities of SUPCRT92, as well as its advanced graphical and numerical capabilities, make R an attractive platform for comprehensive geochemical data analysis, experiment evaluation and modelling. <br>The aim of RedModRphree is to provide the user with an easy-to-use, high-level interface to program algorithms involving geochemical models, which are then solved using the PHREEQC engine: parameter calibration, error and sensitivity analysis, visualization, up to CPU-intensive parallel coupled reactive transport models. Among the out-of-the-box features included in RedModRphree, we highlight the computation and visualization of Pourbaix (Eh-pH) diagrams and the implementation of 1D advective reactive transport supporting the use of surrogate models replacing expensive PHREEQC calculations [4]. RedModRphree is open source and can be installed from https://git.gfz-potsdam.de/delucia/RedModRphree.</p><p><img src="https://contentmanager.copernicus.org/fileStorageProxy.php?f=gnp.72552df75cff57113730161/sdaolpUECMynit/12UGE&app=m&a=0&c=afa6bef86f4c5523b81a86ccdd579fc9&ct=x&pn=gnp.elif&d=1" alt=""></p><p>[1] De Lucia, M. and Kühn, M.: Coupling R and PHREEQC: Efficient Programming of Geochemical Models, Energy Procedia, 40, 464–471, doi.org/10.1016/j.egypro.2013.08.053, 2013.</p><p>[2] Charlton, S.R. and Parkhurst, D.L.: Modules based on the geochemical model PHREEQC for use in scripting and programming languages, Computers & Geosciences 37, 10, 1653–1663, doi.org/10.1016/j.cageo.2011.02.005, 2011.</p><p>[3] Dick, J.M.: CHNOSZ: Thermodynamic Calculations and Diagrams for Geochemistry, Frontiers in Earth Science, 7, https://doi.org/10.3389/feart.2019.00180, 2019.</p><p>[4] Jatnieks, J., De Lucia, M., Dransch, D., and Sips, M.: Data-driven Surrogate Model Approach for Improving the Performance of Reactive Transport Simulations, Energy Procedia, 97, 447–453, doi.org/10.1016/j.egypro.2016.10.047, 2016.</p><p> </p>

The latest developments in Advanced Driver Assistance Systems (ADAS) have greatly enhanced the comfort and safety of drivers. These technologies can identify driver abnormalities like fatigue, inattention, and impairment, which are essential for averting collisions. One of the important aspects of this technology is automated traffic accident detection and prediction, which may help in saving precious human lives. This study aims to explore critical features related to traffic accident detection and prevention. A public US traffic accident dataset was used for the aforementioned task, where various machine learning (ML) models were applied to predict traffic accidents. These ML models included Random Forest, AdaBoost, KNN, and SVM. The models were compared for their accuracies, where Random Forest was found to be the best-performing model, providing the most accurate and reliable classification of accident-related data. Owing to the black box nature of ML models, this best-fit ML model was executed with explainable AI (XAI) methods such as LIME and permutation importance to understand its decision-making for the given classification task. The unique aspect of this study is the introduction of explainable artificial intelligence which enables us to have human-interpretable awareness of how ML models operate. It provides information about the inner workings of the model and directs the improvement of feature engineering for traffic accident detection, which is more accurate and dependable. The analysis identified critical features, including sources, descriptions of weather conditions, time of day (weather timestamp, start time, end time), distance, crossing, and traffic signals, as significant predictors of the probability of an accident occurring. Future ADAS technology development is anticipated to be greatly impacted by the study’s conclusions. A model can be adjusted for different driving scenarios by identifying the most important features and comprehending their dynamics to make sure that ADAS systems are precise, reliable, and suitable for real-world circumstances.

The widespread use of machine learning (ML) models for decision-making raises critical concerns about transparency and accountability – to which an increasingly popular solution is ‘Explainable AI’ (XAI). Here, the object of explanation is technically complex models which are difficult or even impossible to explain. In contrast, this paper makes a call to de-centre models as the object of explanation and look towards the network of ‘machine learning practice’ that bring models into being, and use. We explore this term through an ethnographic study, conducted in collaboration with a large financial services company. Drawing on recent STS research, we ask: what would an explanation look like from a position that recognises the emergent and relational nature of machine learning practice, and how might this contribute to greater accountability and responsibility for ML in use? Inspired by the engaged programme in STS, we explore if and how approaching explanation through ML practice can be mobilised to intervene in how explanations are done in organisations. Our empirical analysis shows an ‘ecology’ of multiple, situated and intra-acting explanations for machine learning practice across a range of human and non-human actors in the company. We argue that while XAI is inevitably partial and limited, its value lies in establishing explanations as an imperative in contexts where ML is implicated in decision-making. Overall, our research suggests a need to widen and deepen the search for explanations and explore the opportunities for provisional, relational and collective interrogations over what can (and can’t) be explained about ML practice.

With the sustainable development of the construction industry, recycled aggregate (RA) has been widely used in concrete preparation to reduce the environmental impact of construction waste. Compressive strength is an essential measure of the performance of recycled aggregate concrete (RAC). In order to understand the correspondence between relevant factors and the compressive strength of recycled concrete and accurately predict the compressive strength of RAC, this paper establishes a model for predicting the compressive strength of RAC using machine learning and hyperparameter optimization techniques. RAC experimental data from published literature as the dataset, extreme gradient boosting (XGBoost), random forest (RF), K-nearest neighbour (KNN), support vector machine regression Support Vector Regression (SVR), and gradient boosted decision tree (GBDT) RAC compressive strength prediction models were developed. The models were validated and compared using correlation coefficients (R2), Root Mean Square Error (RMSE), mean absolute error (MAE), and the gap between the experimental results of the predicted outcomes. In particular, The effects of different hyperparameter optimization techniques (Grid search, Random search, Bayesian optimization-Tree-structured Parzen Estimator, Bayesian optimization- Gaussian Process Regression) on model prediction efficiency and prediction accuracy were investigated. The results show that the optimal combination of hyperparameters can be searched in the shortest time using the Bayesian optimization algorithm based on TPE (Tree-structured Parzen Estimator); the BO-TPE-GBDT RAC compressive strength prediction model has higher prediction accuracy and generalisation ability. This high-performance compressive strength prediction model provides a basis for RAC’s research and practice and a new way to predict the performance of RAC.