Learning Predictive Models Research Articles

GSOR06 Presentation Time: 11:55 AM: Dosimetric Analysis and Machine Learning Predictive Modeling of Local Control in Cervical Cancer

GSOR06 Presentation Time: 11:55 AM: Dosimetric Analysis and Machine Learning Predictive Modeling of Local Control in Cervical Cancer

Practical guide to SHAP analysis: Explaining supervised machine learning model predictions in drug development.

Despite increasing interest in using Artificial Intelligence (AI) and Machine Learning (ML) models for drug development, effectively interpreting their predictions remains a challenge, which limits their impact on clinical decisions. We address this issue by providing a practical guide to SHapley Additive exPlanations (SHAP), a popular feature-based interpretability method, which can be seamlessly integrated into supervised ML models to gain a deeper understanding of their predictions, thereby enhancing their transparency and trustworthiness. This tutorial focuses on the application of SHAP analysis to standard ML black-box models for regression and classification problems. We provide an overview of various visualization plots and their interpretation, available software for implementing SHAP, and highlight best practices, as well as special considerations, when dealing with binary endpoints and time-series models. To enhance the reader's understanding for the method, we also apply it to inherently explainable regression models. Finally, we discuss the limitations and ongoing advancements aimed at tackling the current drawbacks of the method.

A Novel Deep Learning Model for Stock Prediction Integrating Historical Prices, Core Volatility Trends and Financial News

Accurate stock price forecasting is crucial for investors and analysts aiming to make informed decisions and mitigate risks in financial markets. While deep learning models have proven effective in identifying complex patterns in quantitative data, they often struggle to account for non-quantitative factors such as corporate events, market sentiment, and regulatory changes that can cause unpredictable stock price movements. To address this challenge, an ensemble model is introduced that integrates historical price data, financial news, and stock fundamentals to predict next-day stock prices. The model combines one-dimensional Convolutional Neural Networks (1D CNNs), Long Short-Term Memory networks (LSTMs), and Gated Recurrent Units (GRUs), applied independently to each data stream and concurrently as a unified ensemble, to capture both short-term and long-term market behaviors. NVIDIA (NVDA) stock was selected for testing due to its high volatility and rapid price fluctuations, presenting a challenging case for predictive modeling. Using data spanning from April 2014 to March 2024, the model achieved a coefficient of determination (R²) of 0.983 and a mean absolute error (MAE) of $12.72. These results suggest that the ensemble approach effectively captures both quantitative and qualitative factors, making it a promising tool for improving stock price prediction in volatile market conditions.

Large language models to facilitate pregnancy prediction after invitro fertilization.

We evaluated the efficacy of large language models (LLMs), specifically, generative pre-trained transformer-4 (GPT-4), in predicting pregnancy following invitro fertilization (IVF) treatment and compared its accuracy with results from an original published study. Our findings revealed that GPT-4 can autonomously develop and refine advanced machine learning models for pregnancy prediction with minimal human intervention. The prediction accuracy was 0.79, and the area under the receiver operating characteristic curve (AUROC) was 0.89, exceeding or being at least equivalent to the metrics reported in the original study, that is, 0.78 for accuracy and 0.87 for AUROC. The results suggest that LLMs can facilitate data processing, optimize machine learning models in predicting IVF success rates, and provide data interpretation methods. This capacity can help bridge the knowledge gap between data scientists and medical personnel to solve the most pressing clinical challenges. However, more experiments on diverse and larger datasets are needed to validate and promote broader applications of LLMs in assisted reproduction.

Development of machine learning models for prediction of left ventricular systolic dysfunction in patients with isolated left ventricular dilation

Abstract Background/Introduction Isolated left ventricular dilation (ILVD), in the absence of concomitant LV systolic dysfunction, is a risk factor for new reduction in LV ejection fraction (LVEF) and adverse clinical outcomes. We assessed the performance of different machine learning (ML) models in predicting new reduction in LVEF in patients with ILVD. Methods A retrospective analysis of de-identified patient records in the Tempus Database with at least two transthoracic echocardiograms (TTE) conducted between January 2019 and December 2022 were retrospectively analyzed, with the follow up period ending in January 2023. Inclusion/exclusion criteria are shown in Figure 1A. Briefly, patients with ILVD were defined based on echocardiographic parameters as those with LVEF > 50% and LV end diastolic volume indexed (LVEDVi) > 62 mL/m2 for females and > 75 mL/m2 for males, without prior LVEF < 50% and/or grade 2/3 diastolic dysfunction. The endpoint was LVEF < 40% on future (≥90 days) TTE. Five hard voting ML models and one soft voting ensemble model were trained with data parsed from TTE reports (Figure 1B). Feature importance (SHAP) was calculated, and receiver operating characteristic (ROC) and Kaplan Meier curves evaluated the final model (Figure 1C-E). Results A total of 2,962 patients met the inclusion criteria. The median time from index to last TTE was 578 (IQR = 500) days. LVEF < 40% on follow up TTE was noted in 6% of patients (177 patients, median 478 days to endpoint, IQR = 535 days). Of the six models developed, the ensemble model had the highest ROC area under the curve (0.791, Figure 1D), precision (0.163), F1-score (0.266), and specificity (0.773). The ensemble model also had the lowest predicted prevalence (25.5%), although this was higher than the observed prevalence in the test set (5.7%). Patients predicted to develop LVEF < 40% had a 8.2x higher risk of developing LVEF < 40% within five years of index (p<0.001, Figure 1E). Feature importance showed that male gender, higher heart rate, lower LVEF at index TTE, higher LVEDVi, and lower tricuspid annular plane systolic excursion had the highest mean absolute impact on the model’s output probability (Figure 1C). Conclusion Using only data parsed from the index TTE, ensembling a model for the prediction of developing LVEF < 40% in patients with ILVD and without a history of LV diastolic or systolic dysfunction performed the best. Additional analyses including clinical variables such as comorbidities, symptoms, and medications may improve the model’s performance.

Comprehensive machine learning models for prediction of heart failure in 476,393 women and men from the UK Biobank reveal sex differences and underutilized risk factors

Abstract Background Machine learning (ML) models provide potential advantage over ‘traditional’ regression models in heart failure (HF) prediction. Objective To compare performances of Cox PH models and ML survival models for incident HF in men and women without prevalent ischemic heart disease (IHD). We also aimed to identify potential high-risk precursors otherwise ignored by conventional survival models, and to investigate differences between sex-specific models. Methods We included 476,393 participants (55.6% women) from the UK Biobank, after excluding participants with a history of HF or IHD, and defined sex-specific datasets. We predicted incident HF events using over 400 baseline characteristics. We constructed multivariable Cox PH models, which included all predictor variables and subsequently only those remaining after LASSO stability selection. We also developed two supervised ML models (Random Survival Forest (RSF), eXtreme Gradient Survival Boosting (XGBoost)). We identified the 15 most important sex-specific predictors in each model and performances were compared using the C-index. Models were validated using hold-out sets. Results During 12.3 ± 1.9 years of follow-up, 4680 (1.76%) women and 6631 (3.14%) men developed incident HF. XGBoost showed the best performance during model training (C-index, training: 0.89 in men, 0.97 in women; validation 0.77 in men, 0.80 in women). The multivariable Cox model performed second-best (C-index, training: 0.78 in men, 0.82 in women; validation: 0.76 in men, 0.78 in women). RSF performed slightly worse (C-index, training: 0.75 in men, 0.79 in women; validation: 0.75 in men, 0.79 in women) but did not show performance drop during validation. LASSO stability selection performed similar to RSF. Age, self-reported lifetime treatments and medications, cystatin-C, waist circumference and FEV1-scores were identified as strong risk factors in all models for both sexes. Reduced albumin levels and elevated HbA1c were more strongly associated with high risk in men, while elevated systolic BP showed higher importance in women. Traditional Cox models observed CRP as important only in men, while the ML models identified CRP as important for both sexes. Neutrophil count was considered a strong risk factor in both sexes in the traditional Cox models, yet it was not among the most important predictors in both ML models. Presence of other heart disease (which included a.o. pericardial disease, valve disorders and arrhythmias) was an important predictor variable only in the ML models. Conclusion ML models showed similar performance to Cox PH models for HF prediction. Despite this, differences in predictor importance were identified between models. Sex-specific risk predictors were found, and FEV1 score, which is not commonly included in existing models, was identified as an important risk factor. These results suggest that ML models may reveal additional insights that would otherwise remain unnoticed.

Multi-modality deep learning model for prediction of chronic obstructive coronary artery disease

Abstract Background/Introduction The incidence of coronary artery disease (CAD) is on the rise with an increasing burden of stable ischemic heart disease (SIHD) and high prevalence of obesity, diabetes, hyperlipidemia, and other cardiovascular risk factors. Pre-test probability estimation for CAD based on demographic features and characterization of chest pain is insufficient to accurately assess whether further evaluation for CAD is warranted. Purpose To develop and validate a deep learning (DL) algorithm utilizing electrocardiogram (ECG) waveforms and clinical features to predict obstructive CAD. Methods Clinical data from subjects undergoing invasive angiography over a 4-year period at a quaternary care center exclusive of subjects who had concern for acute coronary syndromes were used for model training. ECGs obtained up to 5 years prior to angiography were included; ECGs were partitioned 8:1:1 for training, validation, and test datasets. Obstructive CAD was defined as receiving percutaneous coronary intervention during elective angiography. Convolutional neural network models were developed for ECG waveforms alone (DL-ECG), clinical features from standard risk scores (DL-Clinical), and the combination of ECG waveforms and clinical features (DL-ECG+Clinical); a commonly used pre-test probability estimation tool derived from the CAD Consortium study was used as a comparison (PTP). The models were evaluated based on area under the curve receiver operating characteristics (AUC). Shapley additive explanations (SHAP) analysis was performed to understand feature importance in the DL models. Results A total of 40,602 ECGs and 8,361 coronary angiograms from 7,201 patients were included in the analysis. The median age was 66 (IQR 56-75), 71.9% were men, 34.9% had hypertension, 30.6% had diabetes mellitus, and 39.0% had dyslipidemia. The PTP model (AUC 0.733 [0.717-0.750]) had similar performance as the DL-Clinical model (AUC 0.762 [0.746-0.778]). The DL-ECG model (AUC 0.741 [0.726-0.758]) had similar performance as both the clinical feature models. The DL-ECG+Clinical model (AUC 0.807 [0.793-0.822]) trained on ECG waveforms and clinical features had a superior performance (Figure 1). SHAP analysis for the DL-ECG+Clinical model revealed that ECG waveforms had the largest influence on the model’s output followed by age and chest pain type (Figure 2). Conclusions Multi-modality DL model utilizing ECG and commonly used features can improve prediction of obstructive CAD in SIHD compared to traditional pre-test probability estimates. Prospective research is warranted to determine whether incorporation of DL methods to ECG analysis effectively improves diagnosis and outcomes in obstructive CAD.Figure 1.Figure 2.

Machine learning for prediction of incident heart failure and heart failure hospitalisation: multi-cohort development, validation and association with clinical phenotypes

Abstract Background Delayed heart failure (HF) diagnosis is associated with adverse outcomes and costs of hospitalisation.(1,2) Whilst previous tools have been developed to predict incident HF,(3) risk of HF incidence may correlate poorly with risk of HF hospitalisation, which may be more important in targeting diagnostics. Purpose To develop and validate a novel decision support tool for prediction of incident HF and HF hospitalisation, and investigate association of score with HF with reduced ejection fraction (HFrEF) and preserved ejection fraction (HFpEF). Methods We developed the models (FIND-HF) in UK primary care health record data from individuals aged ≥40 years without known HF (Jan 2, 1998 to Feb 28, 2022), randomly divided into training (80%) and testing (20%) datasets. We evaluated logistic regression and supervised machine learning models for prediction of 1- and 5- year HF and HF hospitalisation risk in the testing dataset with internal bootstrap validation. Models were externally validated for HF hospitalisation in data from a Taiwanese university hospital. Association of FIND-HF score with HF phenotypes of HFpEF and HFrEF was assessed in two prospective cohorts, the MATCH registry of patients diagnosed with HFrEF and the FIND-AF cohort of individuals at higher predicted risk of atrial fibrillation who underwent echocardiography. Results Of 565 284 UK individuals (mean age 68.9 (SD 12.2) years, 49.2% women), incidence of HF and HF hospitalisation was 1.1% and 0.5% at 1-year, and 6.6% and 3.4% at 5-years, respectively. Prediction performance was superior for machine learning algorithms compared with logistic regression, with best performance for XGBoost (area under receiver operating characteristic curve (AUROC) at 1 and 5-years: 0.755 and 0.723 for incident HF, and 0.738 and 0.711 for incident HF hospitalization). On external validation in 106,026 individuals in Taiwan (mean age 64.7 (SD 10.6) years, 52.8 % women), HF hospitalisation incidence was 0.71%, and 2.46% at 1- and 5- years, respectively, and discrimination performance was excellent (AUROC at 1- and 5-years: 0.834 and 0.843). When the FIND-HF score was applied to individuals in the MATCH cohort (n = 133, mean age 69.0 (SD 11.7) years, 38.3% women) and FIND-AF cohort (n = 82, mean age 71.6 (SD 7.5) years, 50.0% women), the distribution of score to clinical phenotypes demonstrated a binomial distribution (Figure 1), and likelihood of HFpEF diagnosis correlated with the numeric score (Figure 2). Conclusions The machine learning FIND-HF decision support tool can accurately identify individuals at risk of incident HF and HF hospitalisation, to enable new approaches for early detection and prevention. The variation of score amongst high risk individuals with different HF clinical phenotypes emphasizes the distinct pathophysiological pathways but shared adverse outcome profile across the HF syndrome.Figure 1Figure 2

Machine learning prediction models for individualized technical success of chronic total occlusion revascularization using anatomical input

Abstract Introduction Chronic total occlusion revascularization (CTO PCI) remains a challenging procedure in the realm of percutaneous coronary interventions. Efforts such as implementing the hybrid algorithm for CTO crossing have been made. Interventional techniques are evolving, and complexity scores such as the J-CTO score might get outdated. Machine learning models can be used to overcome this limitation by constantly providing new data and improving the models by continuous data feeding. In this study, we introduced machine learning models to predict individualized technical success of CTO PCI using anatomical coronary information and baseline information. Methods Consecutive patients undergoing percutaneous coronary intervention of chronic total occlusions were enrolled in this study. Angiographic data and lesion characteristics from coronary angiography were extracted. Machine learning models were applied to predict technical success. A total of 271 patients were used to validate and train the models and 50 separate patients for testing. Data was normalized and fed to several classifying machine learning models including decision trees (DT), random forest (RF), extreme gradient boosting (XGB), CatBoost (CB) classifiers, and light gradient boosting (LGB). Finally, accuracy, precision, recall and F1-score of the selected hyperparameters and cross-validation was performed. Results CB was the best performing model with an accuracy of 0.81, precision of 0.83 and recall of 0.98. The F1 score was 0.90. RF was the second-best performing model with accuracy of 0.77, precision of 0.82, recall of 0.92 and F1 score 0.87. The tuned models performed modestly with RF having a low F1 score of 0.44, due to small sample size. In the tuned models. LGB had a higher F1 score and a relatively high accuracy (0.782) and precision (0.508). The feature importance matrix found age, CTO length, reference diameter to have the highest value among the lesion complexity parameters. Discussion: Overall, the normal-run CB was the best model for predicting the technical success after a CTO PCI with a high accuracy. The fine-tuned and cross-validated models performed modestly because of the relatively small, unbalanced dataset and overfitting. However, data showed promising first results in implementing machine learning into predicting interventional outcomes. Further studies like data augmentation and providing a weighted dataset would increase the model performance and provide a very precise statement whether a patient would have increased chances to have a successful PCI based on baseline anatomic information. Conclusion This is one of the first studies to predict technical success of CTO PCI using angiographic baseline parameters using machine learning. Future studies with larger patient cohorts are needed to involve deploying the selected model in larger patient cohorts including external validation and assessing its performance with new, unseen data.Figure 1.Machine learning modelsFigure 2.Feature importance matrix

Unlocking the future: machine learning models for predicting in-hospital mortality after transcatheter aortic valve replacement: a single-arm meta-analysis

Abstract Background Transcatheter aortic valve replacement (TAVR) continues to be increasingly performed. While machine learning has demonstrated utility in predicting adverse outcomes in various cardiovascular procedures, its effectiveness in predicting in-hospital mortality in TAVR remains uncertain. Purpose We aimed to examine the accuracy of machine learning in predicting in-hospital mortality in TAVR. Methods We searched PubMed and Cochrane databases up to February 2024 for observational studies assessing the diagnostic accuracy of machine learning for predicting in-hospital mortality in TAVR. Extracted data included study characteristics, patient demographics, type of predictive model used, and area under the receiver operating characteristic curves (AUC). Predictive accuracy was assessed by pooling AUC of machine-learning models across studies. AUC values ranging from 0.80 to 0.90, 0.70 to 0.80, 0.60 to 0.70, and 0.50 to 0.60 corresponded to excellent, good, fair, and poor discrimination ability, respectively (1). We used a random effects model to estimate the overall effect size with a 95% confidence interval (CI). Heterogeneity was evaluated with the I2 test.We conducted subgroup analyses based on the type of machine-learning model utilized to assess for any potential differences in predictive accuracy. Results A total of three studies were identified, comprising a study population of 70,714 patients. Across studies, eight machine learning models were utilized, consisting of three artificial neural network models, one logistic regression model, one naive Bayes model, two random forest models, and one support vector machine model. Pooled AUC showed excellent accuracy of machine learning in predicting in-hospital mortality in TAVR (0.92; 95% CI 0.90 - 0.95). Subgroup analyses revealed similar yet excellent accuracy across all types of machine learning models utilized: artificial neural network (0.92; 95% CI 0.86-0.99), logistic regression (0.92; 95% CI 0.89-0.95), naive Bayes (0.90; 95% CI 0.88-0.92), random forest (0.94; 95% CI 0.87-1.00), and support vector machine (0.94; 95% CI 0.91-0.96). Conclusion Our study suggests that utilizing machine learning can accurately predict in-hospital mortality in TAVR. Operators of TAVR procedures may benefit from incorporating machine learning prediction models in their clinical practice. Further exploration of machine learning on outcomes in TAVR is needed to verify our findings.

Detecting Unrecognized Dementia Using Deep Learning Methods with Korean National Claims Data

Abstract Background Dementia is one of the leading causes of death in elderly people. In Korea, the government supports dementia screening for people aged 60 and older, but it is not a population-based organized screening program. It is difficult to achieve early diagnosis and early treatment. The objective of this study is to detect unrecognized dementia using Korean National Claims Data for early diagnosis. Methods This study used the National Health Insurance Claims data in Korea. The case group is the new dementia visit as an unrecognized group, while the control group was randomly selected from the general population without dementia. The predictors included health utilization, socioeconomic and demographic, procedure codes, diagnostic codes and health screenings data. The health utilization variables included length of stay, the number of outpatient visits, health expenditures and other encounter information. All medical utilization data was aggregated monthly. In the case group, 12 months of data from 6-month or 2-years before the onset of dementia were used. In the control group, 12 months of data from randomly selected time point were included. Transformer, embedding methods, and Time2Vec were used as the deep learning methods. This study included 453,306 incidents of dementia, representing almost all cases reported from 2010 to 2019 in Korea. The control group consisted of 669,873. Results After training the deep learning prediction model, the AUC for 6-month dementia prediction on the test dataset was 0.87, and for 2-year dementia prediction, it was 0.76. The most important variables were health utilization variables, especially health expenditures. The health utilization trajectories of the case group were significantly different from those of the control group. Conclusions Based on the claims data, we could detect unrecognized dementia case. We can extend the model to other chronic disease such as cancer and stroke, which are significant burden of disease. Key messages • Screening in healthcare facilities or using lab test results would be more accurate, but it would be difficult to apply to a large population due to the cost-effectiveness and timeliness constraints. • On the other hand, the detection model using the existing claims data would be helpful to to develop cost-effective mass screening program. It can cover the weakness of current screening program.

SLB - SMOTE logistic blending hybrid machine learning model for chronic polycystic ovary syndrome prediction with correlated feature selection

ABSTRACT Objective In this study, we aimed to develop a machine learning (ML) model for predicting Polycystic Ovary Syndrome (PCOS) based on demographic, clinical, and biochemical parameters. Methodology We collected data from Kaggle, which included information on age, body mass index, menstrual cycle length, follicle-stimulating hormone, hair growth, and more. Using this data, we trained several traditional ML and ensemble algorithms to predict PCOS. Results Among the traditional ML algorithms, Logistic Regression emerged as the best, boasting the highest accuracy of 0.91 and an AUC of 0.90. In ensemble algorithms, the Blending algorithm outperformed other ensemble methods, also achieving an accuracy of 0.91 and an AUC of 0.90, with a balanced precision and recall of 0.88. Significance of the Research These results establish Logistic Regression and the Blending algorithm as optimal choices for accurate and reliable PCOS prediction, demonstrating strong discriminative power and the ability to correctly classify PCOS cases.

Evaluating Machine Learning Models for Stroke Prognosis and Prediction in Atrial Fibrillation Patients: A Comprehensive Meta-Analysis

Background/Objective: Atrial fibrillation (AF) complicates the management of acute ischemic stroke (AIS), necessitating precise predictive models to enhance clinical outcomes. This meta-analysis evaluates the efficacy of machine learning (ML) models in three key areas: stroke prognosis in AF patients, stroke prediction in AF patients, and AF prediction in stroke patients. The study aims to assess the accuracy and variability of ML models in forecasting AIS outcomes and detecting AF in stroke patients, while exploring the clinical benefits and limitations of integrating these models into practice. Methods: We conducted a systematic search of PubMed, Embase, and Cochrane databases up to June 2024, selecting studies that evaluated ML accuracy in stroke prognosis and prediction in AF patients and AF prediction in stroke patients. Data extraction and quality assessment were performed independently by two reviewers, with random-effects modeling applied to estimate pooled accuracy metrics. Results: The meta-analysis included twenty-four studies comprising 7,391,645 patients, categorized into groups for stroke prognosis in AF patients (eight studies), stroke prediction in AF patients (thirteen studies), and AF prediction in stroke patients (three studies). The pooled AUROC was 0.79 for stroke prognosis and 0.68 for stroke prediction in AF, with higher accuracy noted in short-term predictions. The mean AUROC across studies was 0.75, with models such as Extreme Gradient Boosting (XGB) and Random Forest (RF) showing superior performance. For stroke prognosis in AF, the mean AUROC was 0.78, whereas stroke prediction yielded a mean AUROC of 0.73. AF prediction post-stroke had an average AUROC of 0.75. These findings indicate moderate predictive capability of ML models, underscoring the need for further refinement and standardization. The absence of comprehensive sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) metrics limited the ability to conduct full meta-analytic modeling. Conclusions: While ML models demonstrate potential for enhancing stroke prognosis and AF prediction, they have yet to meet the clinical standards required for widespread adoption. Future efforts should focus on refining these models and validating them across diverse populations to improve their clinical utility.

Enhanced Diabetes Detection and Blood Glucose Prediction Using TinyML-Integrated E-Nose and Breath Analysis: A Novel Approach Combining Synthetic and Real-World Data

Diabetes mellitus, a chronic condition affecting millions worldwide, necessitates continuous monitoring of blood glucose level (BGL). The increasing prevalence of diabetes has driven the development of non-invasive methods, such as electronic noses (e-noses), for analyzing exhaled breath and detecting biomarkers in volatile organic compounds (VOCs). Effective machine learning models require extensive patient data to ensure accurate BGL predictions, but previous studies have been limited by small sample sizes. This study addresses this limitation by employing conditional generative adversarial networks (CTGAN) to generate synthetic data from real-world tests involving 29 healthy and 29 diabetic participants, resulting in over 14,000 new synthetic samples. These data were used to validate machine learning models for diabetes detection and BGL prediction, integrated into a Tiny Machine Learning (TinyML) e-nose system for real-time analysis. The proposed models achieved an 86% accuracy in BGL identification using LightGBM (Light Gradient Boosting Machine) and a 94.14% accuracy in diabetes detection using Random Forest. These results demonstrate the efficacy of enhancing machine learning models with both real and synthetic data, particularly in non-invasive systems integrating e-noses with TinyML. This study signifies a major advancement in non-invasive diabetes monitoring, underscoring the transformative potential of TinyML-powered e-nose systems in healthcare applications.

Author Correction: Building machine learning prediction models for well-being using predictors from the exposome and genome in a population cohort

Author Correction: Building machine learning prediction models for well-being using predictors from the exposome and genome in a population cohort

Multi-objective optimization of automotive seat frames using machine learning

The optimal design of automobile seats plays an important role in passenger safety in high-speed accidents. In order to enhance the accuracy of the prediction of the input variables and output response of the seat, a hybrid machine learning prediction model that combines the improved gray wolf optimizer (IGWO) and back propagation neural network (BPNN) has been proposed, and the prediction effect of the model was validated using the seat simulation data. Initially, based on the experimental data, finite element models were developed for eight typical working conditions of automobile seats and their accuracy was validated. Subsequently, the energy absorption to mass ratio method was employed to screen the design variables, resulting in the selection of 17 thickness variables and 15 material variables. Thereafter, the gray wolf optimizer (GWO) algorithm underwent enhancement through the incorporation of the dynamic leadership hierarchy (DLH) mechanism and the revision of the positional formula, yielding the IGWO algorithm. Following this, the IGWO algorithm was applied to optimize the hyperparameters of BPNN, culminating in the establishment of the IGWO-BPNN model. Ultimately, the seat multi-objective optimization design process was addressed using multi-objective gray wolf optimizer (MOGWO) to achieve the Pareto frontier, while the decision-making was conducted using the combined compromise solution (CoCoSo) method to determine the best trade-off solution. Furthermore, the effectiveness of the proposed optimal design method is evidenced by comparing the baseline design, simulation analysis, and optimal design methods. The results indicate that the optimized automotive seat frame achieves a reduction in cost by 20.7 % and mass by 22.9 %, simultaneously maintaining safety performance. Consequently, the proposed optimization design methodology is demonstrated to be highly effective for the multi-objective optimization design of automotive seat frames.

Identifying cardiovascular disease risk in the U.S. population using environmental volatile organic compounds exposure: A machine learning predictive model based on the SHAP methodology

BackgroundCardiovascular disease (CVD) remains a leading cause of mortality globally. Environmental pollutants, specifically volatile organic compounds (VOCs), have been identified as significant risk factors. This study aims to develop a machine learning (ML) model to predict CVD risk based on VOC exposure and demographic data using SHapley Additive exPlanations (SHAP) for interpretability. MethodsWe utilized data from the National Health and Nutrition Examination Survey (NHANES) from 2011 to 2018, comprising 5098 participants. VOC exposure was assessed through 15 urinary metabolite metrics. The dataset was split into a training set (70 %) and a test set (30 %). Six ML models were developed, including Random Forest (RF), Light Gradient Boosting Machine (LightGBM), Decision Tree (DT), Extreme Gradient Boosting (XGBoost), Multi-Layer Perceptron (MLP), and Support Vector Machines (SVM). Model performance was evaluated using the Area Under the Receiver Operating Characteristic Curve (AUROC), accuracy, balanced accuracy, F1 score, J-index, kappa, Matthew's correlation coefficient (MCC), positive predictive value (PPV), negative predictive value (NPV), sensitivity (sens), specificity (spec) and SHAP was applied to interpret the best-performing model. ResultsThe RF model exhibited the highest predictive performance with an ROC of 0.8143. SHAP analysis identified age and ATCA as the most significant predictors, with ATCA showing a protective effect against CVD, particularly in older adults and those with hypertension. The study found a significant interaction between ATCA levels and age, indicating that the protective effect of ATCA is more pronounced in older individuals due to increased oxidative stress and inflammatory responses associated with aging. E-values analysis suggested robustness to unmeasured confounders. ConclusionsThis study is the first to utilize VOC exposure data to construct an ML model for predicting CVD risk. The findings highlight the potential of combining environmental exposure data with demographic information to enhance CVD risk prediction, supporting the development of personalized prevention and intervention strategies.

Correction: A spatio-temporal predictive learning model for efficient sea surface temperature forecasting

Correction: A spatio-temporal predictive learning model for efficient sea surface temperature forecasting

Machine learning predictive models and risk factors for lymph node metastasis in non-small cell lung cancer

BackgroundThe prognosis of non-small cell lung cancer (NSCLC) is substantially affected by lymph node metastasis (LNM), but there are no noninvasive, inexpensive methods of relatively high accuracy available to predict LNM in NSCLC patients.MethodsClinical data on NSCLC patients were obtained from the Surveillance, Epidemiology, and End Results (SEER) database. Risk factors for LNM were recognized LASSO and multivariate logistic regression. Six predictive models were constructed with machine learning based on risk factors. The area under the receiver operating characteristic curve (AUC) was used to assess the performance of the model. Subgroup analysis with different T-stages was performed on an optimal model. A webpage LNM risk calculator for optimal model was built using the Shinyapps.io platform.ResultsWe enrolled 64,012 NSCLC patients, of whom 26,611 (41.57%) had LNM. Using multivariate logistic regression, we finally identified 10 independent risk factors for LNM: age, sex, race, histology, primary site, grade, T stage, M stage, tumor size, and bone metastases. GLM is the optimal model among all six machine learning models in both the training and validation cohorts. Subgroup analyses revealed that GLM has good predictability for populations with different T staging. A webpage LNM risk calculator based on GLM was posted on the shinyapps.io platform (https://wubopredict.shinyapps.io/dynnomapp/).ConclusionThe predictive model based on GLM can be used to precisely predict the probability of LNM in NSCLC patients, which was proven effective in all subgroup analyses according to T staging.

Interpretable multi-step hybrid deep learning model for karst spring discharge prediction: Integrating temporal fusion transformers with ensemble empirical mode decomposition

Karst groundwater is a critical freshwater resource for numerous regions worldwide. Monitoring and predicting karst spring discharge is essential for effective groundwater management and the preservation of karst ecosystems. However, the high heterogeneity and karstification pose significant challenges to physics-based models in providing robust predictions of karst spring discharge. In this study, an interpretable multi-step hybrid deep learning model called selective EEMD-TFT is proposed, which adaptively integrates temporal fusion transformers (TFT) with ensemble empirical mode decomposition (EEMD) for predicting karst spring discharge. The selective EEMD-TFT hybrid model leverages the strengths of both EEMD and TFT techniques to learn inherent patterns and temporal dynamics from nonlinear and nonstationary signals, eliminate redundant components, and emphasize useful characteristics of input variables, leading to the improvement of prediction performance and efficiency. It consists of two stages: in the first stage, the daily precipitation data is decomposed into multiple intrinsic mode functions using EEMD to extract valuable information from nonlinear and nonstationary signals. All decomposed components, temperature and categorical date features are then fed into the TFT model, which is an attention-based deep learning model that combines high-performance multi-horizon prediction and interpretable insights into temporal dynamics. The importance of input variables will be quantified and ranked. In the second stage, the decomposed precipitation components with high importance are selected to serve as the TFT model’s input features along with temperature and categorical date variables for the final prediction. Results indicate that the selective EEMD-TFT model outperforms other sequence-to-sequence deep learning models, such as LSTM and single TFT models, delivering reliable and robust prediction performance. Notably, it maintains more consistent prediction performance at longer forecast horizons compared to other sequence-to-sequence models, highlighting its capacity to learn complex patterns from the input data and efficiently extract valuable information for karst spring prediction. An interpretable analysis of the selective EEMD-TFT model is conducted to gain insights into relationships among various hydrological processes and analyze temporal patterns.