Predicting Extraction Selectivity of Acetic Acid in Pervaporation by Machine Learning Models with Data Leakage Management

Interpretable machine learning models to predict decline in intrinsic capacity among older adults in China: a prospective cohort study

BACKGROUND

Monitoring intrinsic capacity and implementing appropriate interventions can support healthy aging. There are, though, few tools available for predicting decline in intrinsic capacity among older adults. This study aimed to develop and validate an interpretable machine learning model designed to identify populations at elevated risk of a decline in intrinsic capacity.

METHODS

Using data from the China Health and Retirement Longitudinal Study baseline (2011) and 4-year follow-up (2015), a total of 822 participants were randomly allocated to a training set and a testing set at a 7:3 ratio. Five machine learning methods were employed to train the model and assess its performance through various metrics. The SHapley Additive exPlanation method was subsequently used to interpret the optimal model.

RESULTS

The 4-year incidence of decline in intrinsic capacity among the older adults in the sample was 44.6 % (n = 367). Nine variables were screened for model construction, among which eXtreme gradient boosting demonstrated the best predictive performance, achieving an area under the receiver operating characteristic curve (AUC) of 0.715 (95 % CI 0.651-0.780) in the testing set. The SHapley Additive exPlanation method identified educational level, smoking, handgrip strength, self-rated health, and residence as the top five significant predictors.

CONCLUSIONS

The developed model can serve as a highly effective tool for primary care teams to identify older adults with early signs of decline in intrinsic capacity, facilitating the provision of subsequent screening and tailored interventions for intrinsic capacity.

Model-driven high-throughput zebrafish embryo assay for evaluating whole effluent toxicity variation across 100 full-scale wastewater treatment plants

The zebrafish embryo is a valuable model for evaluating whole effluent toxicity (WET). However, the widely recognized acute toxicity indicator, based on International Organization of Standardization (ISO) methods, requires large numbers of embryos and is often time-consuming due to its complex experimental procedures. In this study, we propose an alternative to the conventional reliance on ISO standards by developing a model-driven high-throughput assay that utilizes actual wastewater, enabling rapid LC10 (the lethal concentration at which 10 % of the test organisms are affected) prediction through machine learning techniques and multidimensional indicators derived from streamlined experimental procedures. We compared three streamlined toxicity assays-developmental toxicity, behavioral toxicity, and vascular toxicity-along with five different models. Among these, the Lasso model based on behavioral toxicity emerged as the most effective, achieving an R2 value of 0.893 while reducing experimental time by 5- to 8-fold. Furthermore, fivefold cross-validation confirmed its robust predictive accuracy. The application of this model-driven high-throughput assay across 100 wastewater treatment plants in China highlights the crucial role of biological treatment, particularly aerobic processes and secondary sedimentation, in reducing toxicity, thereby providing valuable insights into their functions. This high-throughput assay not only surpasses the ISO standard method in efficiency but also substantially decreases embryo usage, facilitating rapid WET assessments of actual wastewater with larger sample sizes.

Machine Learning Accelerated Discovery of Covalent Organic Frameworks for Environmental and Energy Applications

Covalent organic frameworks (COFs) are porous crystalline materials obtained by linking organic ligands covalently. Their high surface area and adjustable pore sizes make them ideal for a range of applications, including CO2 capture, CH4 storage, gas separation, catalysis, etc. Traditional methods of material research, which mainly rely on manual experimentation, are not particularly efficient, while with advancements in computer science, high-throughput computational screening methods based on molecular simulation have become crucial in material discovery, yet they face limitations in terms of computational resources and time. Currently, machine learning (ML) has emerged as a transformative tool in many fields, capable of analyzing large data sets, identifying underlying patterns, and predicting material performance efficiently and accurately. This approach, termed "materials genomics", combines high-throughput computational screening with ML to predict and design high-performance materials, significantly speeding up the discovery process compared to traditional methods. This review discusses the functions of ML in the screening, design, and performance prediction of COFs and highlights their applications across various domains like CO2 capture, CH4 storage, gas separation, and catalysis, thereby providing new research directions and enhancing the understanding of COF materials and their applications.

Deciphering and Mitigating of Dynamic Greenhouse Gas Emission in Urban Drainage Systems with Knowledge-Infused Graph Neural Network

Deciphering and mitigating dynamic greenhouse gas (GHG) emissions under environmental fluctuation in urban drainage systems (UDGSs) is challenging due to the absence of a high-prediction model that accurately quantifies the contributions of biological production pathways. Here we infused biological production pathways into the graph neural network (GNN) model architecture, developing ecological knowledge-infused GNN (EcoGNN-GHG) models to evaluate methane (CH4) and nitrous oxide (N2O) production in sewers and wastewater treatment plants (WWTPs). The EcoGNN-GHG model demonstrated high predictive accuracy, achieving an R2 of 0.96 for CH4 in sewers and 0.82 for N2O in WWTPs. Model interpretability analysis revealed fluctuations in contributions of the anaerobic hydrolysis acidification pathway to CH4 production and the nitrification-denitrification pathway to N2O production under dynamic environmental conditions, guiding the formulation of a precise dissolved oxygen control strategy targeting critical water quality parameters (acetate for CH4 production and nitrite for N2O production). Implementing this strategy to control DO thereby regulating biological production pathway contributions, CH4 production in sewers and N2O production in WWTPs were reduced by 35.50% and 29.94%, respectively. Our findings offer a robust, accurate method for predicting GHG emissions, quantifying production pathway contributions, and developing effective control strategies in UDGSs.

Transfer Learning with a Graph Attention Network and Weighted Loss Function for Screening of Persistent, Bioaccumulative, Mobile, and Toxic Chemicals

In silico methods for screening hazardous chemicals are necessary for sound management. Persistent, bioaccumulative, mobile, and toxic (PBMT) chemicals persist in the environment and have high mobility in aquatic environments, posing risks to human and ecological health. However, lack of experimental data for the vast number of chemicals hinders identification of PBMT chemicals. Through an extensive search of measured chemical mobility data, as well as persistent, bioaccumulative, and toxic (PBT) chemical inventories, this study constructed comprehensive data sets on PBMT chemicals. To address the limited volume of the PBMT chemical data set, a transfer learning (TL) framework based on graph attention network (GAT) architecture was developed to construct models for screening PBMT chemicals, designating the PBT chemical inventories as source domains and the PBMT chemical data set as target domains. A weighted loss (LW) function was proposed and proved to mitigate the negative impact of imbalanced data on the model performance. Results indicate the TL-GAT models outperformed GAT models, along with large coverage of applicability domains and interpretability. The constructed models were employed to identify PBMT chemicals from inventories consisting of about 1 × 106 chemicals. The developed TL-GAT framework with the LW function holds broad applicability across diverse tasks, especially those involving small and imbalanced data sets.

Perspectives on Advancing Multimodal Learning in Environmental Science and Engineering Studies

The environment faces increasing anthropogenic impacts, resulting in a rapid increase in environmental issues that undermine the natural capital essential for human wellbeing. These issues are complex and often influenced by various factors represented by data with different modalities. While machine learning (ML) provides data-driven tools for addressing the environmental issues, the current ML models in environmental science and engineering (ES&E) often neglect the utilization of multimodal data. With the advancement in deep learning, multimodal learning (MML) holds promise for comprehensive descriptions of the environmental issues by harnessing data from diverse modalities. This advancement has the potential to significantly elevate the accuracy and robustness of prediction models in ES&E studies, providing enhanced solutions for various environmental modeling tasks. This perspective summarizes MML methodologies and proposes potential applications of MML models in ES&E studies, including environmental quality assessment, prediction of chemical hazards, and optimization of pollution control techniques. Additionally, we discuss the challenges associated with implementing MML in ES&E and propose future research directions in this domain.

Optimization of a Novel Engineered Ecosystem Integrating Carbon, Nitrogen, Phosphorus, and Sulfur Biotransformation for Saline Wastewater Treatment Using an Interpretable Machine Learning Approach

The denitrifying sulfur (S) conversion-associated enhanced biological phosphorus removal (DS-EBPR) process for treating saline wastewater is characterized by its unique microbial ecology that integrates carbon (C), nitrogen (N), phosphorus (P), and S biotransformation. However, operational instability arises due to the numerous parameters and intricates bacterial interactions. This study introduces a two-stage interpretable machine learning approach to predict S conversion-driven P removal efficiency and optimize DS-EBPR process. Stage one utilized the XGBoost regression model, achieving an R2 value of 0.948 for predicting sulfate reduction (SR) intensity from anaerobic parameters with feature engineering. Stage two involved the CatBoost classification and regression model integrating anoxic parameters with the predicted SR values for predicting P removal, reaching an accuracy of 94% and an R2 value of 0.93, respectively. This study identified key environmental factors, including SR intensity (20-45 mg S/L), influent P concentration (<9.0 mg P/L), mixed liquor volatile suspended solids (MLVSS)/mixed liquor suspended solids (MLSS) ratio (0.55-0.72), influent C/S ratio (0.5-1.0), anoxic reaction time (5-6 h), and MLSS concentration (>6.50 g/L). A user-friendly graphic interface was developed to facilitate easier optimization and control. This approach streamlines the determination of optimal conditions for enhancing P removal in the DS-EBPR process.

Development of Machine Learning Models for Ion-Selective Electrode Cation Sensor Design

Polyvinyl chloride (PVC) membrane-based ion-selective electrode (ISE) sensors are common tools for water assessments, but their development relies on time-consuming and costly experimental investigations. To address this challenge, this study combines machine learning (ML), Morgan fingerprint, and Bayesian optimization technologies with experimental results to develop high-performance PVC-based ISE cation sensors. By using 1745 data sets collected from 20 years of literature, appropriate ML models are trained to enable accurate prediction and a deep understanding of the relationship between ISE components and sensor performance (R 2 = 0.75). Rapid ionophore screening is achieved using the Morgan fingerprint based on atomic groups derived from ML model interpretation. Bayesian optimization is then applied to identify optimal combinations of ISE materials with the potential to deliver desirable ISE sensor performance. Na+, Mg2+, and Al3+ sensors fabricated from Bayesian optimization results exhibit excellent Nernst slopes with less than 8.2% deviation from the ideal value and superb detection limits at 10-7 M level based on experimental validation results. This approach can potentially transform sensor development into a more time-efficient, cost-effective, and rational design process, guided by ML-based techniques.

Data-driven prediction on critical mechanical properties of engineered cementitious composites based on machine learning

The present study introduces a novel approach utilizing machine learning techniques to predict the crucial mechanical properties of engineered cementitious composites (ECCs), spanning from typical to exceptionally high strength levels. These properties, including compressive strength, flexural strength, tensile strength, and tensile strain capacity, can not only be predicted but also precisely estimated. The investigation encompassed a meticulous compilation and examination of 1532 datasets sourced from pertinent research. Four machine learning algorithms, linear regression (LR), K nearest neighbors (KNN), random forest (RF), and extreme gradient boosting (XGB), were used to establish the prediction model of ECC mechanical properties and determine the optimal model. The optimal model was utilized to employ SHapley Additive exPlanations (SHAP) for scrutinizing feature importance and conducting an in-depth parametric analysis. Subsequently, a comprehensive control strategy was devised for ECC mechanical properties. This strategy can provide actionable guidance for ECC design, equipping engineers and professionals in civil engineering and material science to make informed decisions throughout their design endeavors. The results show that the RF model demonstrated the highest prediction accuracy for compressive strength and flexural strength, with R2 values of 0.92 and 0.91 on the test set. The XGB model outperformed in predicting tensile strength and tensile strain capacity, with R2 values of 0.87 and 0.80 on the test set, respectively. The prediction of tensile strain capacity was the least accurate. Meanwhile, the MAE of the tensile strain capacity was a mere 0.84%, smaller than the variability (1.77%) of the test results in previous research. Compressive strength and tensile strength demonstrated high sensitivity to variations in both water-cement ratio (W) and water reducer (WR). In contrast, flexural strength exhibited high sensitivity solely to changes in W. Conversely, the sensitivity of tensile strain capacity to input features was moderate and consistent. The mechanical attributes of ECC emerged from the combined effects of multiple positive and negative features. Notably, WR exerted the most significant influence on compressive strength among all features, whereas polyethylene (PE) fiber emerged as the primary driver affecting flexural strength, tensile strength, and tensile strain capacity.

Data-driven approaches linking wastewater and source estimation hazardous waste for environmental management

Industrial enterprises are major sources of contaminants, making their regulation vital for sustainable development. Tracking contaminant generation at the firm-level is challenging due to enterprise heterogeneity and the lack of a universal estimation method. This study addresses the issue by focusing on hazardous waste (HW), which is difficult to monitor automatically. We developed a data-driven methodology to predict HW generation using wastewater big data which is grounded in the availability of this data with widespread application of automatic sensors and the logical assumption that a correlation exists between wastewater and HW generation. We created a generic framework that used representative variables from diverse sectors, exploited a data-balance algorithm to address long-tail data distribution, and incorporated causal discovery to screen features and improve computation efficiency. Our method was tested on 1024 enterprises across 10 sectors in Jiangsu, China, demonstrating high fidelity (R² = 0.87) in predicting HW generation with 4,260,593 daily wastewater data.

Machine Learning for Polymer Design to Enhance Pervaporation-Based Organic Recovery

Pervaporation (PV) is an effective membrane separation process for organic dehydration, recovery, and upgrading. However, it is crucial to improve membrane materials beyond the current permeability-selectivity trade-off. In this research, we introduce machine learning (ML) models to identify high-potential polymers, greatly improving the efficiency and reducing cost compared to conventional trial-and-error approach. We utilized the largest PV data set to date and incorporated polymer fingerprints and features, including membrane structure, operating conditions, and solute properties. Dimensionality reduction, missing data treatment, seed randomness, and data leakage management were employed to ensure model robustness. The optimized LightGBM models achieved RMSE of 0.447 and 0.360 for separation factor and total flux, respectively (logarithmic scale). Screening approximately 1 million hypothetical polymers with ML models resulted in identifying polymers with a predicted permeation separation index >30 and synthetic accessibility score <3.7 for acetic acid extraction. This study demonstrates the promise of ML to accelerate tailored membrane designs.

Coupling machine learning and theoretical models to compare key properties of biochar in adsorption kinetics rate and maximum adsorption capacity for emerging contaminants

Insights into key properties of biochar with a fast adsorption rate and high adsorption capacity are urgent to design biochar as an adsorbent in pollution emergency treatment. Machine learning (ML) incorporating classical theoretical adsorption models was applied to build prediction models for adsorption kinetics rate (i.e., K) and maximum adsorption capacity (i.e., Qm) of emerging contaminants (ECs) on biochar. Results demonstrated that the prediction performance of adaptive boosting algorithm significantly improved after data preprocessing (i.e., log-transformation) in the small unbalanced datasets with R2 of 0.865 and 0.874 for K and Qm, respectively. The surface chemistry, primarily led by ash content of biochar significantly influenced the K, while surface porous structure of biochar showed a dominant role in predicting Qm. An interactive platform was deployed for relevant scientists to predict K and Qm of new biochar for ECs. The research provided practical references for future engineered biochar design for ECs removal.

Critical Analysis of Data Leakage in WiFi CSI-Based Human Action Recognition Using CNNs

WiFi Channel State Information (CSI)-based human action recognition using convolutional neural networks (CNNs) has emerged as a promising approach for non-intrusive activity monitoring. However, the integrity and reliability of the reported performance metrics are susceptible to data leakage, wherein information from the test set inadvertently influences the training process, leading to inflated accuracy rates. In this paper, we conduct a critical analysis of a notable IEEE Sensors Journal study on WiFi CSI-based human action recognition, uncovering instances of data leakage resulting from the absence of subject-based data partitioning. Empirical investigation corroborates the lack of exclusivity of individuals across dataset partitions, underscoring the importance of rigorous data management practices. Furthermore, we demonstrate that employing data partitioning with respect to humans results in significantly lower precision rates than the reported 99.9% precision, highlighting the exaggerated nature of the original findings. Such inflated results could potentially discourage other researchers and impede progress in the field by fostering a sense of complacency.

H2 mediated mixed culture microbial electrosynthesis for high titer acetate production from CO2

Microbial electrosynthesis (MES) converts CO2 into value-added products such as volatile fatty acids (VFAs) with minimal energy use, but low production titer has limited scale-up and commercialization. Mediated electron transfer via H2 on the MES cathode has shown a higher conversion rate than the direct biofilm-based approach, as it is tunable via cathode potential control and accelerates electrosynthesis from CO2. Here we report high acetate titers can be achieved via improved in situ H2 supply by nickel foam decorated carbon felt cathode in mixed community MES systems. Acetate concentration of 12.5 g L-1 was observed in 14 days with nickel-carbon cathode at a poised potential of -0.89 V (vs. standard hydrogen electrode, SHE), which was much higher than cathodes using stainless steel (5.2 g L-1) or carbon felt alone (1.7 g L-1) with the same projected surface area. A higher acetate concentration of 16.0 g L-1 in the cathode was achieved over long-term operation for 32 days, but crossover was observed in batch operation, as additional acetate (5.8 g L-1) was also found in the abiotic anode chamber. We observed the low Faradaic efficiencies in acetate production, attributed to partial H2 utilization for electrosynthesis. The selective acetate production with high titer demonstrated in this study shows the H2-mediated electron transfer with common cathode materials carries good promise in MES development.

Understanding Rejection Mechanisms of Trace Organic Contaminants by Polyamide Membranes via Data-Knowledge Codriven Machine Learning

Data-driven machine learning (ML) provides a promising approach to understanding and predicting the rejection of trace organic contaminants (TrOCs) by polyamide (PA). However, various confounding variables, coupled with data scarcity, restrict the direct application of data-driven ML. In this study, we developed a data-knowledge codriven ML model via domain-knowledge embedding and explored its application in comprehending TrOC rejection by PA membranes. Domain-knowledge embedding enhanced both the predictive performance and the interpretability of the ML model. The contribution of key mechanisms, including size exclusion, charge effect, hydrophobic interaction, etc., that dominate the rejections of the three TrOC categories (neutral hydrophilic, neutral hydrophobic, and charged TrOCs) was quantified. Log D and molecular charge emerge as key factors contributing to the discernible variations in the rejection among the three TrOC categories. Furthermore, we quantitatively compared the TrOC rejection mechanisms between nanofiltration (NF) and reverse osmosis (RO) PA membranes. The charge effect and hydrophobic interactions possessed higher weights for NF to reject TrOCs, while the size exclusion in RO played a more important role. This study demonstrated the effectiveness of the data-knowledge codriven ML method in understanding TrOC rejection by PA membranes, providing a methodology to formulate a strategy for targeted TrOC removal.

Feature Engineering and Supervised Machine Learning to Forecast Biogas Production during Municipal Anaerobic Co-Digestion

Municipalities with excess anaerobic digestion capacity accept offsite wastes for co-digestion to meet sustainability goals and create more biogas. Despite the benefits inherent to co-digestion, the temporal and compositional heterogeneity of external waste streams creates operational challenges that lead to upsets or conservative co-digestion. Given the complex microbial bioprocesses occurring during anaerobic digestion, prediction and modeling of the outcomes can be challenging, and machine learning has the potential to improve understanding and control of co-digestion processes. Biogas flows are a surrogate for process health, and here, we predicted biogas production from historical data collected by a water resource recovery facility (WRRF) during normal operation. We tested a daily lab and operational data set (n = 1089 after cleaning) and a minute-by-minute supervisory control and data acquisition (SCADA) operational data set (n = 491,761 after cleaning) to determine if forecasting biogas flow for a 24 h time horizon is feasible without collecting additional data. We found that a multilayer perceptron (MLP) neural network model outperformed tree-based and multiple linear regression models. Using a high-resolution SCADA data set for the first time, we showed that MLP neural networks could predict biogas production with an adjusted coefficient of determination (R2) of 0.78 and a mean absolute percentage error of 13.4% on a holdout test set. Adding daily laboratory analyses to the model did not appreciably improve the prediction of biogas flows. Feature engineering was essential to an accurate prediction, and 11 of the 15 most important features in the SCADA model were calculated from raw SCADA outputs. In summary, this paper demonstrates that minute-scale SCADA information collected at a municipal co-digestion facility can forecast biogas production, as a first step toward a digital twin model, without additional data collection.

Quantifying source contributions to ambient NH3 using Geo-AI with time lag and parcel tracking functions

Ambient ammonia (NH3) plays an important compound in forming particulate matters (PMs), and therefore, it is crucial to comprehend NH3's properties in order to better reduce PMs. However, it is not easy to achieve this goal due to the limited range/real-time NH3 data monitored by the air quality stations. While there were other studies to predict NH3 and its source apportionment, this manuscript provides a novel method (i.e., GEO-AI)) to look into NH3 predictions and their contribution sources. This study represents a pioneering effort in the application of a novel geospatial-artificial intelligence (Geo-AI) base model with parcel tracking functions. This innovative approach seamlessly integrates various machine learning algorithms and geographic predictor variables to estimate NH3 concentrations, marking the first instance of such a comprehensive methodology. The Shapley additive explanation (SHAP) was used to further analyze source contribution of NH3 with domain knowledge. From 2016 to 2018, Taichung's hourly average NH3 values were predicted with total variance up to 96%. SHAP values revealed that waterbody, traffic and agriculture emissions were the most significant factors to affect NH3 concentrations in Taichung among all the characteristics. Our methodology is a vital first step for shaping future policies and regulations and is adaptable to regions with limited monitoring sites.

End-to-end machine-learning for high-gravity ammonia stripping: Bridging the gap between scientific research and user-friendly applications

The removal and recovery of ammonia from wastewater are critical processes for achieving global environmental sustainability and promoting circular economic development. High-gravity technology is an advanced solution to achieve ammonia stripping from wastewater. This study used machine-learning (ML) techniques to provide more comprehensive insights on various influencing factors, including the operating parameters, wastewater characteristics, and design parameters of rotating packed beds. Bayesian auto-optimization combined with a boosting algorithm effectively overcame the challenges of modeling complex datasets with small sample sizes, multidimensional data, missing values, and skewed distributions. Accurate ML based predictive models for the ammonia removal efficiency (η) and mass transfer coefficient (KLa) were developed, the performance on the training set was R2 = 0.98 and R2 = 0.89, and on the testing set was R2 = 0.98 and R2 = 0.82. The developed model revealed that the stripping stage and gas-liquid ratio were the most influential features for predicting η, whereas the liquid flow and high-gravity factor were the most important features for predicting KLa. The well-trained model was then deployed in an online software application that could provide both predictive and auto-update functions for operators and managers, ensuring that practitioners could use the model. The end-to-end machine-learning approach used in this study-that is, covering data collection, model development, and application-could improve the availability of research results, providing valuable references for the further advancement of technology in the field of environmental.

A convenient machine learning model to predict full stomach and evaluate the safety and comfort improvements of preoperative oral carbohydrate in patients undergoing elective painless gastrointestinal endoscopy

BACKGROUND AND AIMS

Assessment of the patient's gastric contents is the key to avoiding aspiration incidents, however, there is no effective method to determine whether elective painless gastrointestinal endoscopy (GIE) patients have a full stomach or an empty stomach. And previous studies have shown that preoperative oral carbohydrates (POCs) can improve the discomfort induced by fasting, but there are different perspectives on their safety. This study aimed to develop a convenient, accurate machine learning (ML) model to predict full stomach. And based on the model outcomes, evaluate the safety and comfort improvements of POCs in empty- and full stomach groups.

METHODS

We enrolled 1386 painless GIE patients between October 2022 and January 2023 in Nanjing First Hospital, and 1090 patients without POCs were used to construct five different ML models to identify full stomach. The metrics of discrimination and calibration validated the robustness of the models. For the best-performance model, we further interpreted it through SHapley Additive exPlanations (SHAP) and constructed a web calculator to facilitate clinical use. We evaluated the safety and comfort improvements of POCs by propensity score matching (PSM) in the two groups, respectively.

RESULTS

Random Forest (RF) model showed the greatest discrimination with the area under the receiver operating characteristic curve (AUROC) 0.837 [95% confidence interval (CI): 79.1-88.2], F1 71.5%, and best calibration with a Brier score of 15.2%. The web calculator can be visited at https://medication.shinyapps.io/RF_model/. PSM results demonstrated that POCs significantly reduced the full stomach incident in empty stomach group (p < 0.05), but no differences in full stomach group (p > 0.05). Comfort improved in both groups and was more significant in empty stomach group.

CONCLUSIONS

The developed convenient RF model predicted full stomach with high accuracy and interpretability. POCs were safe and comfortably improved in both groups, with more benefit in empty stomach group. These findings may guide the patients' gastrointestinal preparation.

Machine Learning in Environmental Research: Common Pitfalls and Best Practices

Machine learning (ML) is increasingly used in environmental research to process large data sets and decipher complex relationships between system variables. However, due to the lack of familiarity and methodological rigor, inadequate ML studies may lead to spurious conclusions. In this study, we synthesized literature analysis with our own experience and provided a tutorial-like compilation of common pitfalls along with best practice guidelines for environmental ML research. We identified more than 30 key items and provided evidence-based data analysis based on 148 highly cited research articles to exhibit the misconceptions of terminologies, proper sample size and feature size, data enrichment and feature selection, randomness assessment, data leakage management, data splitting, method selection and comparison, model optimization and evaluation, and model explainability and causality. By analyzing good examples on supervised learning and reference modeling paradigms, we hope to help researchers adopt more rigorous data preprocessing and model development standards for more accurate, robust, and practicable model uses in environmental research and applications.

Machine Learning-Assisted Design of Thin-Film Composite Membranes for Solvent Recovery

Organic solvents are extensively utilized in industries as raw materials, reaction media, and cleaning agents. It is crucial to efficiently recover solvents for environmental protection and sustainable manufacturing. Recently, organic solvent nanofiltration (OSN) has emerged as an energy-efficient membrane technology for solvent recovery; however, current OSN membranes are largely fabricated by trial-and-error methods. In this study, for the first time, we develop a machine learning (ML) approach to design new thin-film composite membranes for solvent recovery. The monomers used in interfacial polymerization, along with membrane, solvent and solute properties, are featurized to train ML models via gradient boosting regression. The ML models demonstrate high accuracy in predicting OSN performance including solvent permeance and solute rejection. Subsequently, 167 new membranes are designed from 40 monomers and their OSN performance is predicted by the ML models for common solvents (methanol, acetone, dimethylformamide, and n-hexane). New top-performing membranes are identified with methanol permeance superior to that of existing membranes. Particularly, nitrogen-containing heterocyclic monomers are found to enhance microporosity and contribute to higher permeance. Finally, one new membrane is experimentally synthesized and tested to validate the ML predictions. Based on the chemical structures of monomers, the ML approach developed here provides a bottom-up strategy toward the rational design of new membranes for high-performance solvent recovery and many other technologically important applications.