Machine Learning-based Prediction Model Research Articles

Construction of a machine learning-based prediction model for unfavourable discharge outcomes in patients with ischemic stroke

Construction of a machine learning-based prediction model for unfavourable discharge outcomes in patients with ischemic stroke

Predicting all-cause mortality and premature death using interpretable machine learning among a middle-aged and elderly Chinese population

ObjectiveTo develop machine learning-based prediction models for all-cause and premature mortality among the middle-aged and elderly population in China. MethodAdults aged 45 years or older at baseline of 2011 from the China Health and Retirement Longitudinal Study (CHARLS) were included. The stacked ensemble model was built utilizing five selected machine learning algorithms. These models underwent training and testing using the CHARLS 2011–2015 cohort (derivation cohort) and subsequently underwent external validation using the CHARLS 2015–2018 cohort (validation cohort). SHapley Additive exPlanations (SHAP) was introduced to quantify the importance of risk factors and explain machine learning algorithms. ResultIn derivation cohort, a total of 10,677 subjects were included, 478 died during the follow-up. The stacked ensemble model demonstrated the highest efficacy in terms of its discrimination capability for predicting all-cause mortality and premature death, with an AUC[95 % CI] of 0.826[0.792–0.859] and 0.773[0.725–0.821], respectively. In validation cohort, the corresponding AUC[95 % CI] were 0.803[0.743–0.864] and 0.791[0.719–0.863], respectively. Risk factors including age, sex, self-reported health, activities of daily living, cognitive function, ever smoker, levels of systolic blood pressure, Cystatin C and low density lipoprotein were strong predictors for both all-cause mortality and premature death. ConclusionStacked ensemble models performed well in predicting all-cause and premature death in this Chinese cohort. Interpretable techniques can aid in identifying significant risk factors and non-linear relationships between predictors and mortality.

A machine learning-based prediction model for delayed clinically important postoperative nausea and vomiting in high-risk patients undergoing laparoscopic gastrointestinal surgery

BackgroundDelayed clinically important postoperative nausea and vomiting (CIPONV) could lead to significant consequences following surgery. We aimed to develop a prediction model for it using machine learning algorithms utilizing perioperative data from patients undergoing laparoscopic gastrointestinal surgery. MethodsAll 1154 patients in the FDP-PONV trial were enrolled. The optimal features for model development were selected by least absolute shrinkage and selection operator and stepwise regression from 81 perioperative variables. The machine learning algorithm with the best area under the receiver operating characteristic curve (ROCAUC) was determined and assessed. The interpretation of the prediction model was performed by the SHapley Additive Explanations library. ResultsSix important predictors were identified. The random forest model showed the best performance in predicting delayed CIPONV, achieving an ROCAUC of 0.737 in the validation cohort. ConclusionThis study developed an interpretable model predicting personalized risk for delayed CIPONV, aiding high-risk patient identification and prevention strategies.

Machine Learning-Based Prediction Models for Punching Shear Strength of Fiber-Reinforced Polymer Reinforced Concrete Slabs Using a Gradient-Boosted Regression Tree.

Fiber-reinforced polymers (FRPs) are increasingly being used as a composite material in concrete slabs due to their high strength-to-weight ratio and resistance to corrosion. However, FRP-reinforced concrete slabs, similar to traditional systems, are susceptible to punching shear failure, a critical design concern. Existing empirical models and design provisions for predicting the punching shear strength of FRP-reinforced concrete slabs often exhibit significant bias and dispersion. These errors highlight the need for more reliable predictive models. This study aims to develop gradient-boosted regression tree (GBRT) models to accurately predict the shear strength of FRP-reinforced concrete panels and to address the limitations of existing empirical models. A comprehensive database of 238 sets of experimental results for FRP-reinforced concrete slabs has been compiled from the literature. Different machine learning algorithms were considered, and the performance of GBRT models was evaluated against these algorithms. The dataset was divided into training and testing sets to verify the accuracy of the model. The results indicated that the GBRT model achieved the highest prediction accuracy, with root mean square error (RMSE) of 64.85, mean absolute error (MAE) of 42.89, and coefficient of determination (R2) of 0.955. Comparative analysis with existing experimental models showed that the GBRT model outperformed these traditional approaches. The SHapley Additive exPlanation (SHAP) method was used to interpret the GBRT model, providing insight into the contribution of each input variable to the prediction of punching shear strength. The analysis emphasized the importance of variables such as slab thickness, FRP reinforcement ratio, and critical section perimeter. This study demonstrates the effectiveness of the GBRT model in predicting the punching shear strength of FRP-reinforced concrete slabs with high accuracy. SHAP analysis elucidates key factors that influence model predictions and provides valuable insights for future research and design improvements.

A machine learning-based prediction model for architectural heritage: the case of domed Sinan mosques

This study presents a machine learning-based prediction model (PM) customized to predict missing components of historical mosques. Domed mosques built by Architect Sinan during the Classical Ottoman Period (16th century) are selected due to their distinctive features and stylistic similarities. The model development process includes data collection (46 domed Sinan mosques), data preparation and refinement, training, testing, and validation. The Pix2Pix method is used to train and validate the machine learning models, and the Structural Similarity (SSIM) metric is used to objectively evaluate the outcomes. Preliminary results indicate that the success of the PMs is not directly proportional to the number of input components. Instead, factors such as overall mass organization, the curvature of the dome, and the number of balconies on the minaret play crucial roles in determining the success of the outcomes.

Construction and verification of a machine learning-based prediction model of deep vein thrombosis formation after spinal surgery

Construction and verification of a machine learning-based prediction model of deep vein thrombosis formation after spinal surgery

Predicting compressive strength of hollow concrete prisms using machine learning techniques and explainable artificial intelligence (XAI).

The design of masonry structures requires accurate estimation of compressive strength (CS) of hollow concrete masonry prisms. Generally, the CS of masonry prisms is determined by destructive laboratory testing which results in time and resource wastage. Thus, this study aims to provide machine learning-based predictive models for CS of hollow concrete masonry blocks using different algorithms including Multi Expression Programming (MEP), Random Forest Regression (RFR), and Extreme Gradient Boosting (XGB) etc. A dataset of 159 experimental results was collected from published literature for this purpose. The collected dataset consisted of four input parameters including strength of masonry units (), height-to-thickness ratio (h/t), strength of mortar () , and ratio of and only one output parameter i.e., CS. Out of all the algorithms employed in current study, only MEP and GEP expressed their output in the form of an empirical equation. The accuracy of developed models was assessed using root mean squared error (RMSE), objective function (OF), and etc. Among all algorithms assessed, XGB turned out to be the most accurate having = 0.99 and least OF value of 0.0063 followed by AdaBoost, RFR, and other algorithms. The developed XGB model was also used to conduct different explainable artificial intelligence (XAI) analysis including sensitivity and shapley analysis and the results showed that strength of masonry unit () is the most significant variable in predicting CS. Thus, the ML-based predictive models presented in this study can be utilized practically for determining CS of hollow concrete masonry prisms without requiring expensive and time-consuming laboratory testing.

Machine Learning-Based Prediction Model for the Yield of Nitrogen-Enriched Biomass Pyrolysis Products: Performance Evaluation and Interpretability Analysis

The process optimization and control of nitrogen-rich biomass pyrolysis technology is crucial. This technology effectively converts agricultural waste into high-value energy products. The aim of this study is to develop an accurate model for predicting the yield of nitrogen-rich biomass pyrolysis products using machine learning techniques. The goal is to improve both the yield and quality of the products. This study uses machine learning techniques to develop an accurate model for predicting the yield of three-phase products from nitrogen-rich biomass pyrolysis. The goal is to optimize the biomass energy conversion process. The study builds a dataset containing 468 samples. This dataset comprehensively characterizes the chemical, elemental, and industrial properties of biomass, as well as the pyrolysis operating parameters. Models with high prediction accuracy are achieved by comparing various machine learning algorithms. These algorithms include Random Forest, GBDT, SVR, and ANN. Fine feature selection and hyper-parameter tuning are used to enhance the performance. In particular, the optimized GBDT model achieves a performance of MSE of 10.04 and R² of 0.92 in cross-validation. SHAP value analysis is used to further interpret the prediction results of the model. Sensitivity analysis, conducted through Monte Carlo simulation, identifies key features that have a high impact on the target value. Additionally, two-feature partial correlation analysis reveals the importance and dependence of the model features. This provides crucial technical support and a scientific basis for the industrial application of biomass energy and the optimization of the pyrolysis process.

Leveraging Greenhouse Gas Emissions Traceability in the Groundnut Supply Chain: Blockchain-Enabled Off-Chain Machine Learning as a Driver of Sustainability

AbstractAs emphasized in multiple United Nations (UN) reports, sustainable agriculture, a key goal in the UN Sustainable Development Goals (SDGs), calls for dedicated efforts and innovative solutions. In this study, greenhouse gas (GHG) emissions in the groundnut supply chain from the region of Diourbel & Niakhar, Senegal, to the port of Dakar are investigated. The groundnut supply chain is divided into three steps: cultivation, harvesting, and processing/shipping. This work adheres to UN guidelines, addressing the imperative for sustainable agriculture by applying machine learning-based predictive modeling (MLPMs) utilizing the FAOSTAT and EDGAR databases. Additionally, it provides a novel approach using blockchain-enabled off-chain machine learning through smart contracts built on Hyperledger Fabric to secure GHG emissions storage and machine learning’s predictive analytics from fraud and enhance transparency and data security. This study also develops a decision-making dashboard to provide actionable insights for GHG emissions reduction strategies across the groundnut supply chain.

Machine learning-based prediction model of lower extremity deep vein thrombosis after stroke.

This study aimed to apply machine learning (ML) techniques to develop and validate a risk prediction model for post-stroke lower extremity deep vein thrombosis (DVT) based on patients' limb function, activities of daily living (ADL), clinical laboratory indicators, and DVT preventive measures. We retrospectively analyzed 620 stroke patients. Eight ML models-logistic regression (LR), support vector machine (SVM), random forest (RF), decision tree (DT), neural network (NN), extreme gradient boosting (XGBoost), Bayesian (NB), and K-nearest neighbor (KNN)-were used to build the model. These models were extensively evaluated using ROC curves, AUC, PR curves, PRAUC, accuracy, sensitivity, specificity, and clinical decision curves (DCA). Shapley's additive explanation (SHAP) was used to determine feature importance. Finally, based on the optimal ML algorithm, different functional feature set models were compared with the Padua scale to select the best feature set model. Our results indicated that the RF algorithm demonstrated superior performance in various evaluation metrics, including AUC (0.74/0.73), PRAUC (0.58/0.58), accuracy (0.75/0.77), and sensitivity (0.78/0.80) in both the training set and test set. DCA analysis revealed that the RF model had the highest clinical net benefit. SHAP analysis showed that D-dimer had the most significant influence on DVT, followed by age, Brunnstrom stage (lower limb), prothrombin time (PT), and mobility ability. The RF algorithm can predict post-stroke DVT to guide clinical practice.

Integrative prediction model for radiation pneumonitis incorporating genetic and clinical-pathological factors using machine learning

PurposeWe aimed to develop a machine learning-based prediction model for severe radiation pneumonitis (RP) by integrating relevant clinicopathological and genetic factors, considering the associations of clinical, dosimetric parameters, and single nucleotide polymorphisms (SNPs) of genes in the TGF-β1 pathway with RP. MethodsWe prospectively enrolled 59 primary lung cancer patients undergoing radiotherapy and analyzed pretreatment blood samples, clinicopathological/dosimetric variables, and 11 functional SNPs in TGFβ pathway genes. Using the Synthetic Minority Over-sampling Technique (SMOTE) and nested cross-validation, we developed a machine learning-based prediction model for severe RP (grade ≥ 2). Feature selection was conducted using four methods (filtered-based, wrapper-based, embedded, and logistic regression), and performance was evaluated using three machine learning models. ResultsSevere RP occurred in 20.3 % of patients with a median follow-up of 39.7 months. In our final model, age (>66 years), smoking history, PTV volume (>300 cc), and AG/GG genotype in BMP2 rs1979855 were identified as the most significant predictors. Additionally, incorporating genomic variables for prediction alongside clinicopathological variables significantly improved the AUC compared to using clinicopathological variables alone (0.822 vs. 0.741, p = 0.029). The same feature set was selected using both the wrapper-based method and logistic model, demonstrating the best performance across all machine learning models (AUC: XGBoost 0.815, RF 0.805, SVM 0.712, respectively). ConclusionWe successfully developed a machine learning-based prediction model for RP, demonstrating age, smoking history, PTV volume, and BMP2 rs1979855 genotype as significant predictors. Notably, incorporating SNP data significantly enhanced predictive performance compared to clinicopathological factors alone.

Designing an Experimental Platform to Assess Ergonomic Factors and Distraction Index in Law Enforcement Vehicles during Mission-Based Routes

Mission-based routes for various occupations play a crucial role in occupational driver safety, with accident causes varying according to specific mission requirements. This study focuses on the development of a system to address driver distraction among law enforcement officers by optimizing the Driver–Vehicle Interface (DVI). Poorly designed DVIs in law enforcement vehicles, often fitted with aftermarket police equipment, can lead to perceptual-motor problems such as obstructed vision, difficulty reaching controls, and operational errors, resulting in driver distraction. To mitigate these issues, we developed a driving simulation platform specifically for law enforcement vehicles. The development process involved the selection and placement of sensors to monitor driver behavior and interaction with equipment. Key criteria for sensor selection included accuracy, reliability, and the ability to integrate seamlessly with existing vehicle systems. Sensor positions were strategically located based on previous ergonomic studies and digital human modeling to ensure comprehensive monitoring without obstructing the driver’s field of view or access to controls. Our system incorporates sensors positioned on the dashboard, steering wheel, and critical control interfaces, providing real-time data on driver interactions with the vehicle equipment. A supervised machine learning-based prediction model was devised to evaluate the driver’s level of distraction. The configured placement and integration of sensors should be further studied to ensure the updated DVI reduces driver distraction and supports safer mission-based driving operations.

Machine Learning-Based Prediction Models of Mortality for Intensive Care Unit Patients Using Nursing Records.

This study aimed to develop ICU mortality prediction models using a conceptual framework, focusing on nurses' concerns reflected in nursing records from the MIMIC IV database. We included 46,693 first-time ICU admissions of adults over 18 years with a minimum 24-hour stay, excluding those receiving hospice or palliative care. Predictors included demographics, clinical characteristics, and nursing documentation frequencies related to nurses' concerns. Four models were trained with 10-fold cross-validation after adjusting class imbalance. The random forest (RF) model was identified as the best-performing, with key predictors of mortality in this model being the frequency of vital signs, the frequency of nursing note documentation, and the frequency of monitoring-related nursing notes. This suggests that predictive models using nursing records, which reflect nurses' concerns as represented by the frequency of nursing documentation, may be integrated into clinical decision support tools, potentially enhancing patient outcomes in ICUs.

Machine learning-based prediction models for electoral outcomes in India: a comparative analysis of exit polls from 2014–2021

Machine learning-based prediction models for electoral outcomes in India: a comparative analysis of exit polls from 2014–2021

Applying machine learning in screening for Down Syndrome in both trimesters for diverse healthcare scenarios

BackgroundThis paper describes the development of low-cost, effective, non-invasive machine learning-based prediction models for Down Syndrome in the first two trimesters of pregnancy in Vietnam. These models are adaptable to different situations with limited screening capacities at community-based healthcare facilities. MethodUltrasound and biochemical testing alone and in combination, from both trimesters were employed to build prediction models based on k-Nearest Neighbor, Support Vector Machine, Random Forest, and Extreme Gradient Boosting algorithms. ResultsA total of 7,076 pregnant women from a single site in Northern Vietnam were included, and 1,035 had a fetus with Down Syndrome. Combined ultrasound and biochemical testing were required to achieve the highest accuracy in trimester 2, while models based only on biochemical testing performed as well as models based on combined testing during trimester 1. In trimester 1, Extreme Gradient Boosting produced the best model with 94% accuracy and 88% AUC, while Support Vector Machine produced the best model in trimester 2 with 89% accuracy and 84% AUC. ConclusionsThis study explored a range of machine learning models under different testing scenarios. Findings point to the potential feasibility of national screening, especially in settings without enough equipment and specialists, after additional model validation and fine tuning is performed.

Identification of risk factors of Long COVID and predictive modeling in the RECOVER EHR cohorts

BackgroundSARS-CoV-2-infected patients may develop new conditions in the period after the acute infection. These conditions, the post-acute sequelae of SARS-CoV-2 infection (PASC, or Long COVID), involve a diverse set of organ systems. Limited studies have investigated the predictability of Long COVID development and its associated risk factors.MethodsIn this retrospective cohort study, we used electronic healthcare records from two large-scale PCORnet clinical research networks, INSIGHT (~1.4 million patients from New York) and OneFlorida+ (~0.7 million patients from Florida), to identify factors associated with having Long COVID, and to develop machine learning-based models for predicting Long COVID development. Both SARS-CoV-2-infected and non-infected adults were analysed during the period of March 2020 to November 2021. Factors associated with Long COVID risk were identified by removing background associations and correcting for multiple tests.ResultsWe observed complex association patterns between baseline factors and a variety of Long COVID conditions, and we highlight that severe acute SARS-CoV-2 infection, being underweight, and having baseline comorbidities (e.g., cancer and cirrhosis) are likely associated with increased risk of developing Long COVID. Several Long COVID conditions, e.g., dementia, malnutrition, chronic obstructive pulmonary disease, heart failure, PASC diagnosis U099, and acute kidney failure are well predicted (C-index > 0.8). Moderately predictable conditions include atelectasis, pulmonary embolism, diabetes, pulmonary fibrosis, and thromboembolic disease (C-index 0.7–0.8). Less predictable conditions include fatigue, anxiety, sleep disorders, and depression (C-index around 0.6).ConclusionsThis observational study suggests that association patterns between investigated factors and Long COVID are complex, and the predictability of different Long COVID conditions varies. However, machine learning-based predictive models can help in identifying patients who are at risk of developing a variety of Long COVID conditions.

Neural reinforcement learning signals predict recovery from impulse control disorder symptoms in Parkinson’s disease

BackgroundImpulse control disorders (ICD) in Parkinson’s disease (PD) are associated with a heavy burden on patients and caretakers. While recovery can occur, ICD persists in many patients despite optimal management. The basis for this inter-individual variability in recovery is unclear and poses a major challenge to personalized health care. MethodsWe adopt a computational psychiatry approach and leverage the longitudinal, prospective Personalized Parkinson Project (N=136 persons with PD, within 5 years of diagnosis) to combine dopaminergic learning theory-informed fMRI with machine learning (at baseline) to predict ICD symptom recovery after two years of follow-up. We focused on a change in QUIP-rs across the entire cohort, regardless of an ICD diagnosis. ResultsGreater reinforcement learning signals during gain trials but not loss trials at baseline, including those in the ventral striatum, medial prefrontal cortex and the behavioral accuracy score measured while ON medication were associated with greater recovery from impulse control symptoms two years later. These signals accounted for a unique proportion of the relevant variability over and above that explained by other known factors, such as decreases in dopamine agonist use. ConclusionsOur results provide a proof of principle for combining generative model-based inference of latent learning processes with machine learning-based predictive modeling of variability in clinical symptom recovery trajectories. Hence, we showed that RL modelling parameters predict recovery from ICD symptoms in PD

INDELpred: Improving the prediction and interpretation of indel pathogenicity within the clinical genome

Small insertions and deletions (InDels) are critical yet challenging genetic variations with significant clinical implications. However, the identification of pathogenic InDels from neutral variants in clinical contexts remains an understudied problem. Here, we developed INDELpred, a machine learning-based predictive model for discerning pathogenic from benign InDels. INDELpred was established based on key features, including allele frequency, InDel length, function-based features, and gene-based features. A set of comprehensive evaluation analyses demonstrated that INDELpred exhibited superior performance over competing methods in terms of computational efficiency and prediction accuracy. Importantly, INDELpred highlighted the crucial role of function-based features in identifying pathogenic InDels, with a clear interpretability of the features in understanding the disease-causing variants. We envisage INDELpred as a desirable tool for the detection of pathogenic InDels within large-scale genomic datasets, and thereby enhancing the precision of genetic diagnoses in clinical settings.

The construction of machine learning-based predictive models for high-quality embryo formation in poor ovarian response patients with progestin-primed ovarian stimulation

ObjectiveTo explore the optimal models for predicting the formation of high-quality embryos in Poor Ovarian Response (POR) Patients with Progestin-Primed Ovarian Stimulation (PPOS) using machine learning algorithms.MethodsA retrospective analysis was conducted on the clinical data of 4,216 POR cycles who underwent in vitro fertilization (IVF) / intracytoplasmic sperm injection (ICSI) at Sichuan Jinxin Xinan Women and Children’s Hospital from January 2015 to December 2021. Based on the presence of high-quality cleavage embryos 72 h post-fertilization, the samples were divided into the high-quality cleavage embryo group (N = 1950) and the non-high-quality cleavage embryo group (N = 2266). Additionally, based on whether high-quality blastocysts were observed following full blastocyst culture, the samples were categorized into the high-quality blastocyst group (N = 124) and the non-high-quality blastocyst group (N = 1800). The factors influencing the formation of high-quality embryos were analyzed using logistic regression. The predictive models based on machine learning methods were constructed and evaluated accordingly.ResultsDifferential analysis revealed that there are statistically significant differences in 14 factors between high-quality and non-high-quality cleavage embryos. Logistic regression analysis identified 14 factors as influential in forming high-quality cleavage embryos. In models excluding three variables (retrieved oocytes, MII oocytes, and 2PN fertilized oocytes), the XGBoost model performed slightly better (AUC = 0.672, 95% CI = 0.636–0.708). Conversely, in models including these three variables, the Random Forest model exhibited the best performance (AUC = 0.788, 95% CI = 0.759–0.818). In the analysis of high-quality blastocysts, significant differences were found in 17 factors. Logistic regression analysis indicated that 13 factors influence the formation of high-quality blastocysts. Including these variables in the predictive model, the XGBoost model showed the highest performance (AUC = 0.813, 95% CI = 0.741–0.884).ConclusionWe developed a predictive model for the formation of high-quality embryos using machine learning methods for patients with POR undergoing treatment with the PPOS protocol. This model can help infertility patients better understand the likelihood of forming high-quality embryos following treatment and help clinicians better understand and predict treatment outcomes, thus facilitating more targeted and effective interventions.

Optimizing pre-copy live virtual machine migration in cloud computing using machine learning-based prediction model

AbstractOne of the preconditions for efficient cloud computing services is the continuous availability of services to clients. However, there are various reasons for temporary service unavailability due to routine maintenance, load balancing, cyber-attacks, power management, fault tolerance, emergency incident response, and resource usage. Live Virtual Machine Migration (LVM) is an option to address service unavailability by moving virtual machines between hosts without disrupting running services. Pre-copy memory migration is a common LVM approach used in cloud systems, but it faces challenges due to the high rate of frequently updated memory pages known as dirty pages. Transferring these dirty pages during pre-copy migration prolongs the overall migration time. If there are large numbers of remaining memory pages after a predefined iteration of page transfer, the stop-and-copy phase is initiated, which significantly increases downtime and negatively impacts service availability. To mitigate this issue, we introduce a prediction-based approach that optimizes the migration process by dynamically halting the iteration phase when the predicted downtime falls below a predefined threshold. Our proposed machine learning method was rigorously evaluated through experiments conducted on a dedicated testbed using KVM/QEMU technology, involving different VM sizes and memory-intensive workloads. A comparative analysis against proposed pre-copy methods and default migration approach reveals a remarkable improvement, with an average 64.91% reduction in downtime for different RAM configurations in high-write-intensive workloads, along with an average reduction in total migration time of approximately 85.81%. These findings underscore the practical advantages of our method in reducing service disruptions during live virtual machine migration in cloud systems.