Development Of Machine Learning Models Research Articles

Learning together: Towards foundation models for machine learning interatomic potentials with meta-learning

The development of machine learning models has led to an abundance of datasets containing quantum mechanical (QM) calculations for molecular and material systems. However, traditional training methods for machine learning models are unable to leverage the plethora of data available as they require that each dataset be generated using the same QM method. Taking machine learning interatomic potentials (MLIPs) as an example, we show that meta-learning techniques, a recent advancement from the machine learning community, can be used to fit multiple levels of QM theory in the same training process. Meta-learning changes the training procedure to learn a representation that can be easily re-trained to new tasks with small amounts of data. We then demonstrate that meta-learning enables simultaneously training to multiple large organic molecule datasets. As a proof of concept, we examine the performance of a MLIP refit to a small drug-like molecule and show that pre-training potentials to multiple levels of theory with meta-learning improves performance. This difference in performance can be seen both in the reduced error and in the improved smoothness of the potential energy surface produced. We therefore show that meta-learning can utilize existing datasets with inconsistent QM levels of theory to produce models that are better at specializing to new datasets. This opens new routes for creating pre-trained, foundation models for interatomic potentials.

Exploring deep fully convolutional neural networks for surface defect detection in complex geometries

AbstractIn this paper, we propose a machine learning approach for detecting superficial defects in metal surfaces using point cloud data. We compare the performance of two popular deep learning architectures, multilayer perceptron networks (MLPs) and fully convolutional networks (FCNs), with varying feature sets. Our results show that FCNs (F1=0.94) outperformed MLPs (F1=0.52) in terms of precision, recall, and F1-score. We found that transfer learning with pre-trained models can improve performance when the amount of available data is limited. Our study highlights the importance of considering the amount and quality of training data in developing machine learning models for defect detection in industrial settings with 3D images.

Machine Learned Classification of Ligand Intrinsic Activities at Human μ-Opioid Receptor.

Opioids are small-molecule agonists of μ-opioid receptor (μOR), while reversal agents such as naloxone are antagonists of μOR. Here, we developed machine learning (ML) models to classify the intrinsic activities of ligands at the human μOR based on the SMILES strings and two-dimensional molecular descriptors. We first manually curated a database of 983 small molecules with measured Emax values at the human μOR. Analysis of the chemical space allowed identification of dominant scaffolds and structurally similar agonists and antagonists. Decision tree models and directed message passing neural networks (MPNNs) were then trained to classify agonistic and antagonistic ligands. The hold-out test AUCs (areas under the receiver operator curves) of the extra-tree (ET) and MPNN models are 91.5 ± 3.9% and 91.8 ± 4.4%, respectively. To overcome the challenge of a small data set, a student-teacher learning method called tritraining with disagreement was tested using an unlabeled data set comprised of 15,816 ligands of human, mouse, and rat μOR, κOR, and δOR. We found that the tritraining scheme was able to increase the hold-out AUC of MPNN models to as high as 95.7%. Our work demonstrates the feasibility of developing ML models to accurately predict the intrinsic activities of μOR ligands, even with limited data. We envisage potential applications of these models in evaluating uncharacterized substances for public safety risks and discovering new therapeutic agents to counteract opioid overdoses.

Development and validation of machine learning models for predicting HER2-zero and HER2-low breast cancers.

To develop and validate machine learning models for HER2-zero and -low using MRI features pre-neoadjuvant therapy (pre-NAT). 516 breast cancer patients post-NAT surgery were randomly divided into training (n = 362) and internal validation sets (n = 154) for model building and evaluation. MRI features (tumor diameter, enhancement type, background parenchymal enhancement, enhancement pattern, percentage of enhancement, signal enhancement ratio, breast edema, and ADC) were reviewed. Logistic regression (LR), support vector machine (SVM), k-nearest neighbor (KNN), and extreme gradient boosting (XGBoost) models utilized MRI characteristics for HER2 status assessment in training and validation datasets. The best-performing model generated a HER2 score, subsequently correlated with pathological complete response (pCR) and disease-free survival (DFS). The XGBoost model outperformed LR, SVM, and KNN, achieving an area under the ROC curve (AUC) of 0.783(95% CI: 0.733-0.833) and 0.787(95% CI: 0.709-0.865) in the validation dataset. Its HER2 score for predicting pCR had an AUC of 0.708 in the training datasets and 0.695 in the validation dataset. Additionally, the low HER2 score was significantly associated with shorter DFS in the validation dataset (HR: 2.748,95% CI: 1.016-7.432, P = 0.037). The XGBoost model could help distinguish HER2-zero and HER2-low breast cancers, and has the potential to predict pCR and prognosis in breast cancer patients undergoing NAT. Human epidermal growth factor receptor 2 (HER2)-low-expressing breast cancer can benefit from the HER2 targeted therapy. Prediction of HER2-low expression is crucial for appropriate management. MRI features offer a solution to this clinical issue.

Meta-Analysis and Machine Learning Models for Anaerobic Biodegradation Rates of Organic Contaminants in Sediments and Sludge.

Anaerobic biodegradation rates (half-lives) of organic chemicals are pivotal for environmental risk assessment and remediation. Traditional experimental evaluation, constrained by prolonged, oxygen-free conditions, struggles to keep pace with emerging contaminants. Data-driven machine learning (ML) models serve as promising complements. However, reported quantitative structure-biodegradation relationships or ML models on anaerobic biodegradation are mostly based on small data sets (<100 records) and neglect experimental conditions, usually achieving compromised predictions. This work aimed to develop ML models for predicting the biodegradation half-lives of organic pollutants in anaerobic environments (i.e., sediment/soil and sludge). Focusing on important features of both chemicals and experimental conditions, we first curated two data sets, one for sediment/soil (SED) and the other for sludge (SLD), covering 978 records for 206 chemicals from the literature, and then conducted a meta-analysis. Next, we built a binary classification (half-life of 30 days as the cutoff) model with an accuracy of 81% and a regression model with R2 of 0.56 for SED based on LightGBM (80% and 0.31 for SLD based on Extra tree, respectively). The model interpretations underscored the significance of experimental conditions (e.g., temperature and inoculum dosage), as evidenced by their high feature importance, and the models were found to correctly capture the effects of chemical substructures, for example, branched structures and aromatic rings prolonged half-lives while methyl group and ortho-substitution on rings shortened half-lives. The applicability domains of the models were also defined, resulting in reasonable prediction for the half-lives of 41% (SED) or 67% (SLD) of over 4000 persistent, bioaccumulative, and toxic chemicals. Overall, this study pioneers ML models for predicting the anaerobic degradation half-lives, offering valuable support for future evaluation and implementation of chemical anaerobic biodegradation.

Development of an Intelligent Road-Crossing Sensor System for Enhanced Pedestrian Safety

Pedestrian safety at road crossings in urban environments is a critical issue due to the increasing vehicular traffic and high pedestrian density. Traditional traffic control measures often fail to adequately address the dynamic interactions between vehicles and pedestrians, leading to significant risks of accidents. This study proposes the development of an intelligent road-crossing sensor system leveraging advanced technologies such as sensors, artificial intelligence (AI), and the Internet of Things (IoT). The system aims to enhance pedestrian safety by providing real-time alerts and adaptive traffic control, thereby reducing the risk of pedestrian-vehicle collisions. The methodology includes sensor selection, data collection, machine learning model development, and system implementation and testing. The potential benefits and challenges of deploying such systems are also discussed, offering insights into creating smarter and safer urban environments. Keywords: Pedestrian safety, intelligent sensor systems, road crossing, vehicle detection, traffic signals

CALiSol-23: Experimental electrolyte conductivity data for various Li-salts and solvent combinations

Ion transport in non-aqueous electrolytes is crucial for high performance lithium-ion battery (LIB) development. The design of superior electrolytes requires extensive experimentation across the compositional space. To support data driven accelerated electrolyte discovery efforts, we curated and analyzed a large dataset covering a wide range of experimentally recorded ionic conductivities for various combinations of lithium salts, solvents, concentrations, and temperatures. The dataset is named as ’Conductivity Atlas for Lithium salts and Solvents’ (CALiSol-23). Comprehensive datasets are lacking but are critical to building chemistry agnostic machine learning models for conductivity as well as data driven electrolyte optimization tasks. CALiSol-23 was derived from an exhaustive review of literature concerning experimental non-aqueous electrolyte conductivity measurement. The final dataset consists of 13,825 individual data points from 27 different experimental articles, in total covering 38 solvents, a broad temperature range, and 14 lithium salts. CALiSol-23 can help expedite machine learning model development that can help in understanding the complexities of ion transport and streamlining the optimization of non-aqueous electrolyte mixtures.

CALiSol-23: Experimental electrolyte conductivity data for various Li-salts and solvent combinations

Ion transport in non-aqueous electrolytes is crucial for high performance lithium-ion battery (LIB) development. The design of superior electrolytes requires extensive experimentation across the compositional space. To support data driven accelerated electrolyte discovery efforts, we curated and analyzed a large dataset covering a wide range of experimentally recorded ionic conductivities for various combinations of lithium salts, solvents, concentrations, and temperatures. The dataset is named as ’Conductivity Atlas for Lithium salts and Solvents’ (CALiSol-23). Comprehensive datasets are lacking but are critical to building chemistry agnostic machine learning models for conductivity as well as data driven electrolyte optimization tasks. CALiSol-23 was derived from an exhaustive review of literature concerning experimental non-aqueous electrolyte conductivity measurement. The final dataset consists of 13,825 individual data points from 27 different experimental articles, in total covering 38 solvents, a broad temperature range, and 14 lithium salts. CALiSol-23 can help expedite machine learning model development that can help in understanding the complexities of ion transport and streamlining the optimization of non-aqueous electrolyte mixtures.

Machine learning-augmented multi-arrayed fiber bragg grating sensors for enhanced structural health monitoring by discriminating strain and temperature variations

AbstractFiber optic sensors (FOS) in long-term structural health monitoring (SHM) have drawn significant attention due to their pivotal role in detecting defects and measuring structural performance in diverse infrastructures. While using FOS, temperature variation due to environmental factors is still considered one of the major challenges to isolating sensing parameters. To address this issue, we reported a machine learning (ML)-augmented multi-parameter sensing system that enables simultaneous detection of strain and temperature effects based on one single fiber Bragg gratings (FBGs) sensor for SHM. The initial phase entailed designing, fabricating, and characterizing a novel FBG sensor in the laboratory, incorporating a set of four FBGs, each distinguished by distinct Bragg wavelengths. In the next phase, ML algorithms are employed to separate temperature effects from strain variations. As a proof of concept, mechanical loading tests are conducted on the sensor, exposing the FBG portion to various temperature conditions. In the final phase, data collected from a post-tensioned concrete bridge embedded with both strain and temperature FBG sensors are utilized, and the developed ML models are applied to observe real-environment outcomes. Despite the limited feature points of collected FBG spectrums, the developed ML models effectively address cross-sensitivity issues induced by temperature perturbations. The long-term benefit of using FOS is that it will enable a better understanding and utilization of aging infrastructure. This will potentially reduce embodied carbon of infrastructure in the future and assist in the global efforts to achieve Net-Zero.

Development and validation of machine learning models to predict MDRO colonization or infection on ICU admission by using electronic health record data

BackgroundMultidrug-resistant organisms (MDRO) pose a significant threat to public health. Intensive Care Units (ICU), characterized by the extensive use of antimicrobial agents and a high prevalence of bacterial resistance, are hotspots for MDRO proliferation. Timely identification of patients at high risk for MDRO can aid in curbing transmission, enhancing patient outcomes, and maintaining the cleanliness of the ICU environment. This study focused on developing a machine learning (ML) model to identify patients at risk of MDRO during the initial phase of their ICU stay.MethodsUtilizing patient data from the First Medical Center of the People’s Liberation Army General Hospital (PLAGH-ICU) and the Medical Information Mart for Intensive Care (MIMIC-IV), the study analyzed variables within 24 h of ICU admission. Machine learning algorithms were applied to these datasets, emphasizing the early detection of MDRO colonization or infection. Model efficacy was evaluated by the area under the receiver operating characteristics curve (AUROC), alongside internal and external validation sets.ResultsThe study evaluated 3,536 patients in PLAGH-ICU and 34,923 in MIMIC-IV, revealing MDRO prevalence of 11.96% and 8.81%, respectively. Significant differences in ICU and hospital stays, along with mortality rates, were observed between MDRO positive and negative patients. In the temporal validation, the PLAGH-ICU model achieved an AUROC of 0.786 [0.748, 0.825], while the MIMIC-IV model reached 0.744 [0.723, 0.766]. External validation demonstrated reduced model performance across different datasets. Key predictors included biochemical markers and the duration of pre-ICU hospital stay.ConclusionsThe ML models developed in this study demonstrated their capability in early identification of MDRO risks in ICU patients. Continuous refinement and validation in varied clinical contexts remain essential for future applications.

A privacy-preserving platform oriented medical healthcare and its application in identifying patients with candidemia

Federated learning (FL) has emerged as a significant method for developing machine learning models across multiple devices without centralized data collection. Candidemia, a critical but rare disease in ICUs, poses challenges in early detection and treatment. The goal of this study is to develop a privacy-preserving federated learning framework for predicting candidemia in ICU patients. This approach aims to enhance the accuracy of antifungal drug prescriptions and patient outcomes. This study involved the creation of four predictive FL models for candidemia using data from ICU patients across three hospitals in China. The models were designed to prioritize patient privacy while aggregating learnings across different sites. A unique ensemble feature selection strategy was implemented, combining the strengths of XGBoost’s feature importance and statistical test p values. This strategy aimed to optimize the selection of relevant features for accurate predictions. The federated learning models demonstrated significant improvements over locally trained models, with a 9% increase in the area under the curve (AUC) and a 24% rise in true positive ratio (TPR). Notably, the FL models excelled in the combined TPR + TNR metric, which is critical for feature selection in candidemia prediction. The ensemble feature selection method proved more efficient than previous approaches, achieving comparable performance. The study successfully developed a set of federated learning models that significantly enhance the prediction of candidemia in ICU patients. By leveraging a novel feature selection method and maintaining patient privacy, the models provide a robust framework for improved clinical decision-making in the treatment of candidemia.

Predicting hardness of graphene-added Si3N4 using machine learning: A data-driven approach

This study presents a data-driven framework based on machine learning (ML) using extreme gradient boosting (XGBoost) for predicting the hardness of silicon nitride (Si3N4) ceramics reinforced with graphene. The XGBoost model takes into account various factors such as graphene type and content, characteristics of the raw Si3N4 powder, the parameters of the sintering process (sintering technique, temperature, pressure, holding time), and the characteristics of the sintered samples, i.e., the density, αβ content and Vickers hardness. The parameters that influence the Si3N4 hardness most strongly are identified, with sintering pressure, sintering time and density being the most influential. The addition of graphene content up to a certain threshold (1 wt%) has a positive impact on hardness. However, beyond that it leads to a lower density and a lower mechanical performance. Sintering parameters, particularly the sintering pressure, temperature, holding time and technique, strongly affect the density, final grain size, αβ Si3N4 composition and subsequently the hardness. The study highlights the importance of density and the densification process in achieving high hardness in Si3N4 ceramics. The developed ML model provides a valuable tool for predicting the hardness of Si3N4+graphene ceramics composites and offers insights into selecting suitable graphene type, content, and processing parameters. While the study primarily focuses on Si3N4+graphene composites, this novel approach holds promise for the in-silico design and analysis of diverse ceramic materials.

Machine Learning for the estimation of foliar nitrogen content in pineapple crops using multispectral images and Internet of Things (IoT) platforms

Nitrogen is the most important nutritional element during the vegetative growth phase of the pineapple crop; however, its presence in the soil is insufficient to meet plant demands. In this study, nine machine learning techniques were validated to estimate the total nitrogen (TN) content in MD2 pineapple crops from data from multiple sources. These sources included multispectral images captured by an unmanned aerial vehicle (UAV); in situ sensors, which collected information on ecological factors such as pH, temperature, solar radiation, relative humidity, soil moisture, wind speed and direction, as well as SPAD values indicating leaf chlorophyll content.Total nitrogen (TN) values were taken from leaf tissue samples, which were then analyzed in a laboratory. To introduce nitrogen variability, a complete randomized block experimental design was implemented, applying five different treatments in five blocks, each with 12 replications, during a 6-month period in a pineapple crop located in Tauramena, Colombia. To address the inherent variability in the agricultural and environmental data, dimensionality was reduced using Principal Component Analysis (PCA). In addition, regularization techniques were applied, including cross-validation, feature selection, boost methods, L1 (Lasso) and L2 (Ridge) regularization, as well as hyperparameter optimization. These strategies generated more robust and accurate models, with the multilayer perceptron regressor (MLP regressor) and extreme gradient boosting (XGBoost) algorithms standing out. On the first sampling date, XGBoost achieved an R2 of 86.98 %, being the highest. On the following dates, MLP achieved a R2 of 59.11 % on the second date; XGBoost achieved a R2 of 68.00 % on the third date, and on the last date, MLP achieved a R2 of 69.4 %. These results indicate that the integration of data from multiple sources and the use of machine learning models could greatly improve the precision of nitro-gen (N) diagnostics in pineapple crops, especially in real-time applications. These findings highlight the promising potential of developing machine learning models that integrate multisensor data fusion for various applications in agriculture.

Gamified Crowdsourcing as a Novel Approach to Lung Ultrasound Data Set Labeling: Prospective Analysis.

Machine learning (ML) models can yield faster and more accurate medical diagnoses; however, developing ML models is limited by a lack of high-quality labeled training data. Crowdsourced labeling is a potential solution but can be constrained by concerns about label quality. This study aims to examine whether a gamified crowdsourcing platform with continuous performance assessment, user feedback, and performance-based incentives could produce expert-quality labels on medical imaging data. In this diagnostic comparison study, 2384 lung ultrasound clips were retrospectively collected from 203 emergency department patients. A total of 6 lung ultrasound experts classified 393 of these clips as having no B-lines, one or more discrete B-lines, or confluent B-lines to create 2 sets of reference standard data sets (195 training clips and 198 test clips). Sets were respectively used to (1) train users on a gamified crowdsourcing platform and (2) compare the concordance of the resulting crowd labels to the concordance of individual experts to reference standards. Crowd opinions were sourced from DiagnosUs (Centaur Labs) iOS app users over 8 days, filtered based on past performance, aggregated using majority rule, and analyzed for label concordance compared with a hold-out test set of expert-labeled clips. The primary outcome was comparing the labeling concordance of collated crowd opinions to trained experts in classifying B-lines on lung ultrasound clips. Our clinical data set included patients with a mean age of 60.0 (SD 19.0) years; 105 (51.7%) patients were female and 114 (56.1%) patients were White. Over the 195 training clips, the expert-consensus label distribution was 114 (58%) no B-lines, 56 (29%) discrete B-lines, and 25 (13%) confluent B-lines. Over the 198 test clips, expert-consensus label distribution was 138 (70%) no B-lines, 36 (18%) discrete B-lines, and 24 (12%) confluent B-lines. In total, 99,238 opinions were collected from 426 unique users. On a test set of 198 clips, the mean labeling concordance of individual experts relative to the reference standard was 85.0% (SE 2.0), compared with 87.9% crowdsourced label concordance (P=.15). When individual experts' opinions were compared with reference standard labels created by majority vote excluding their own opinion, crowd concordance was higher than the mean concordance of individual experts to reference standards (87.4% vs 80.8%, SE 1.6 for expert concordance; P<.001). Clips with discrete B-lines had the most disagreement from both the crowd consensus and individual experts with the expert consensus. Using randomly sampled subsets of crowd opinions, 7 quality-filtered opinions were sufficient to achieve near the maximum crowd concordance. Crowdsourced labels for B-line classification on lung ultrasound clips via a gamified approach achieved expert-level accuracy. This suggests a strategic role for gamified crowdsourcing in efficiently generating labeled image data sets for training ML systems.

Research on the Inversion of Chlorophyll-a Concentration in the Hong Kong Coastal Area Based on Convolutional Neural Networks

Chlorophyll-a (Chl-a) concentration is a key indicator for assessing the eutrophication level in water bodies. However, accurately inverting Chl-a concentrations in optically complex coastal waters presents a significant challenge for traditional models. To address this, we employed Sentinel-2 MSI sensor data and leveraged the power of five machine learning models, including a convolutional neural network (CNN), to enhance the inversion process in the coastal waters near Hong Kong. The CNN model demonstrated superior performance with on-site data validation, outperforming the other four models (R2 = 0.810, RMSE = 1.165 μg/L, MRE = 35.578%). The CNN model was employed to estimate Chl-a concentrations from images captured over the study area in April and October 2022, resulting in the creation of a thematic map illustrating the spatial distribution of Chl-a levels. The map indicated high Chl-a concentrations in the northeast and southwest areas of Hong Kong Island and low Chl-a concentrations in the southeast facing the open sea. Analysis of patch size effects on CNN model accuracy indicated that 7 × 7 and 9 × 9 patches yielded the most optimal results across the tested sizes. Shapley additive explanations were employed to provide post-hoc interpretations for the best-performing CNN model, highlighting that features B6, B12, and B8 were the most important during the inversion process. This study can serve as a reference for developing machine learning models to invert water quality parameters.

Fair prediction of 2-year stroke risk in patients with atrial fibrillation.

This study aims to develop machine learning models that provide both accurate and equitable predictions of 2-year stroke risk for patients with atrial fibrillation across diverse racial groups. Our study utilized structured electronic health records (EHR) data from the All of Us Research Program. Machine learning models (LightGBM) were utilized to capture the relations between stroke risks and the predictors used by the widely recognized CHADS2 and CHA2DS2-VASc scores. We mitigated the racial disparity by creating a representative tuning set, customizing tuning criteria, and setting binary thresholds separately for subgroups. We constructed a hold-out test set that not only supports temporal validation but also includes a larger proportion of Black/African Americans for fairness validation. Compared to the original CHADS2 and CHA2DS2-VASc scores, significant improvements were achieved by modeling their predictors using machine learning models (Area Under the Receiver Operating Characteristic curve from near 0.70 to above 0.80). Furthermore, applying our disparity mitigation strategies can effectively enhance model fairness compared to the conventional cross-validation approach. Modeling CHADS2 and CHA2DS2-VASc risk factors with LightGBM and our disparity mitigation strategies achieved decent discriminative performance and excellent fairness performance. In addition, this approach can provide a complete interpretation of each predictor. These highlight its potential utility in clinical practice. Our research presents a practical example of addressing clinical challenges through the All of Us Research Program data. The disparity mitigation framework we proposed is adaptable across various models and data modalities, demonstrating broad potential in clinical informatics.

Machine learning prediction of methane, nitrogen, and natural gas mixture viscosities under normal and harsh conditions

The accurate estimation of gas viscosity remains a pivotal concern for petroleum engineers, exerting substantial influence on the modeling efficacy of natural gas operations. Due to their time-consuming and costly nature, experimental measurements of gas viscosity are challenging. Data-based machine learning (ML) techniques afford a resourceful and less exhausting substitution, aiding research and industry at gas modeling that is incredible to reach in the laboratory. Statistical approaches were used to analyze the experimental data before applying machine learning. Seven machine learning techniques specifically Linear Regression, random forest (RF), decision trees, gradient boosting, K-nearest neighbors, Nu support vector regression (NuSVR), and artificial neural network (ANN) were applied for the prediction of methane (CH4), nitrogen (N2), and natural gas mixture viscosities. More than 4304 datasets from real experimental data utilizing pressure, temperature, and gas density were employed for developing ML models. Furthermore, three novel correlations have developed for the viscosity of CH4, N2, and composite gas using ANN. Results revealed that models and anticipated correlations predicted methane, nitrogen, and natural gas mixture viscosities with high precision. Results designated that the ANN, RF, and gradient Boosting models have performed better with a coefficient of determination (R2) of 0.99 for testing data sets of methane, nitrogen, and natural gas mixture viscosities. However, linear regression and NuSVR have performed poorly with a coefficient of determination (R2) of 0.07 and − 0.01 respectively for testing data sets of nitrogen viscosity. Such machine learning models offer the industry and research a cost-effective and fast tool for accurately approximating the viscosities of methane, nitrogen, and gas mixture under normal and harsh conditions.

Risk factors and development of machine learning diagnostic models for lateral lymph node metastasis in rectal cancer: multicentre study.

The diagnostic criteria for lateral lymph node metastasis in rectal cancer have not been established. This research aimed to investigate the risk factors for lateral lymph node metastasis and develop machine learning models combining these risk factors to improve the diagnostic performance of standard imaging. This multicentre prospective study included patients who underwent lateral lymph node dissection without preoperative treatment for rectal cancer between 2017 and 2019 in 15 Japanese institutions. First, preoperative clinicopathological factors and magnetic resonance imaging findings were evaluated using multivariable analyses for their correlation with lateral lymph node metastasis. Next, machine learning diagnostic models for lateral lymph node metastasis were developed combining these risk factors. The models were tested in a training set and in an internal validation cohort and their diagnostic performance was tested using receiver operating characteristic curve analyses. Of 212 rectal cancers, 122 patients were selected, including 232 lateral pelvic sides, 30 sides of which had pathological lateral lymph node metastasis. Multivariable analysis revealed that poorly differentiated/mucinous adenocarcinoma, extramural vascular invasion, tumour deposit and a short-axis diameter of lateral lymph node ≥ 6.0 mm were independent risk factors for lateral lymph node metastasis. Patients were randomly divided into a training cohort (139 sides) and a test cohort (93 sides) and machine learning models were computed on the basis of a combination of significant features (including: histological type, extramural vascular invasion, tumour deposit, short- and long-axis diameter of lateral lymph node, body mass index, serum carcinoembryonic antigen level, cT, cN, cM, irregular border and mixed signal intensity). The top three models with the highest sensitivity in the training cohort were as follows: support vector machine (sensitivity, 1.000; specificity, 0.773), light gradient boosting machine (sensitivity, 0.950; specificity, 0.918) and ensemble learning (sensitivity, 0.950; specificity, 0.917). The diagnostic performances of these models in the test cohort were as follows: support vector machine (sensitivity, 0.750; specificity, 0.667), light gradient boosting machine (sensitivity, 0.500; specificity, 0.852) and ensemble learning (sensitivity, 0.667; specificity, 0.864). Machine learning models combining multiple risk factors can contribute to improving diagnostic performance of lateral lymph node metastasis.

Machine Learning Model Development to Predict Power Outage Duration (POD): A Case Study for Electric Utilities.

In the face of increasing climate variability and the complexities of modern power grids, managing power outages in electric utilities has emerged as a critical challenge. This paper introduces a novel predictive model employing machine learning algorithms, including decision tree (DT), random forest (RF), k-nearest neighbors (KNN), and extreme gradient boosting (XGBoost). Leveraging historical sensors-based and non-sensors-based outage data from a Turkish electric utility company, the model demonstrates adaptability to diverse grid structures, considers meteorological and non-meteorological outage causes, and provides real-time feedback to customers to effectively address the problem of power outage duration. Using the XGBoost algorithm with the minimum redundancy maximum relevance (MRMR) feature selection attained 98.433% accuracy in predicting outage durations, better than the state-of-the-art methods showing 85.511% accuracy on average over various datasets, a 12.922% improvement. This paper contributes a practical solution to enhance outage management and customer communication, showcasing the potential of machine learning to transform electric utility responses and improve grid resilience and reliability.

Predicting subacute ruminal acidosis from milk mid-infrared estimated fatty acids and machine learning on Canadian commercial dairy herds

Our objective was to validate the possibility of detecting SARA from milk Fourier transform mid-infrared spectroscopy estimated fatty acids (FA) and machine learning. Subacute ruminal acidosis is a common condition in modern commercial dairy herds for which the diagnostic remains challenging due to its symptoms often being subtle, nonexclusive, and not immediately apparent. This observational study aimed at evaluating the possibility of predicting SARA by developing machine learning models to be applied to farm data and to provide an estimated portrait of SARA prevalence in commercial dairy herds. A first data set composed of 488 milk samples of 67 cows (initial DIM = 8.5 ± 6.18; mean ± SD) from 7 commercial dairy farms and their corresponding SARA classification (SARA+ if rumen pH <6.0 for 300 min, else SARA-) was used for the development of machine learning models. Three sets of predictive variables: i) milk major components (MMC), ii) milk FA (MFA), and iii) MMC combined with MFA (MMCFA) were submitted to 3 different algorithms, namely Elastic net (EN), Extreme gradient boosting (XGB), and Partial least squares (PLS), and evaluated using 3 different scenarios of cross-validation. Accuracy, sensitivity, and specificity of the resulting 27 models were analyzed using a linear mixed model. Model performance was not significantly affected by the choice of algorithm. Model performance was improved by including fatty acids estimations (MFA and MMCFA as opposed to MMC alone). Based on these results, one model was selected (algorithm: EN; predictive variables: MMCFA; 60.4, 65.4, and 55.3% of accuracy, sensitivity, and specificity, respectively) and applied to a large data set comprising the first test-day record (milk major components and FA within the first 70 DIM of 211,972 Holstein cows (219,503 samples) collected from 3001 commercial dairy herds. Based on this analysis, the within-herd SARA prevalence of commercial farms was estimated at 6.6 ± 5.29% ranging from 0 to 38.3%. A subsequent linear mixed model was built to investigate the herd-level factors associated to higher within-herd SARA prevalence. Milking system, proportion of primiparous cows, herd size and seasons were all herd-level factors affecting SARA prevalence. Furthermore, milk production was positively, and milk fat yield negatively associated with SARA prevalence. Due to their moderate levels of accuracy, the SARA prediction models developed in our study, using data from continuous pH measurements on commercial farms, are not suitable for diagnostic purpose. However, these models can provide valuable information at the herd level.