Gradient Boosting Research Articles

Machine learning-based prediction of new onset of atrial fibrillation after mitral valve surgery

Abstract Background New-onset postoperative atrial fibrillation (nPOAF) is a common complication after cardiac surgery (30–50%), being associated with unfavorable long-term outcomes. Using the Society of Thoracic Surgeons National Adult Cardiac Database, we used machine learning (ML) to predict nPOAF and related 30-day outcomes following mitral valve (MV) surgery. A total of 27,856 MV operations were performed at 910 centers between 7/1/2017 and 6/30/2020 on patients without AF or a prior permanent pacemaker. The primary endpoint was nPOAF postoperatively. ML techniques utilized included penalized logistic regression, gradient boosting, decision trees, and random forests. Results The overall incidence of nPOAF was 35.4% and that of new pacemaker insertion was 5.6%. Patients who developed nPOAF were older (67 ± 10 vs 60 ± 13 years), had more mitral valve stenosis (14.1% vs 11.7%), and hypertension (72.1% vs 63.3%). They underwent more mitral valve replacement (39.1% vs 32.7%) and coronary artery bypass grafting (23.9% vs 16%). For predicting nPOAF, ML methods offer sensitivity, specificity and precision superior to logistic regression. The accuracy rate was identical with penalized and non-penalized logistic regression (0.672). Conclusions Predicting nPOAF and its short-term sequelae following MV surgery remains highly challenging. Machine learning methods offer a moderate degree of improvement in predicting nPOAF even in large national-level studies, in the absence of multi-modal data, such as real-time wearables data, electrocardiograms, heart rhythm monitoring, or cardiac imaging.

A highly accurate and robust prediction framework for drilling rate of penetration based on machine learning ensemble algorithm

The rate of penetration (ROP) is a key indicator of drilling efficiency. Many researchers have explored the application of machine learning in ROP prediction. However, few studies have focused on the robustness of the constructed models, and developing a ROP prediction model that can achieve both high accuracy and strong robustness remains a challenge. This paper introduces a novel machine learning approach to constructing a ROP prediction model through ensemble learning algorithms. The model is based on field data from oilfields, incorporating 16 input parameters that influence ROP. First, the feasibility of the collected dataset is verified using correlation analysis. Then, ROP prediction models are developed based on various machine learning algorithms, including Decision Tree Regression (DTR), Random Forest (RF), eXtreme Gradient Boosting (XGB), Light Gradient Boosting Machine (LGBM), Support Vector Regression (SVR), and Backpropagation Neural Network (BPNN). By comparing the performance of these models under noise levels of 0%, 1.7%, and 5.1%, RF, LGBM, XGB, and SVR are selected as base learners. These base learners are then combined to construct multiple ensemble models, and the performance of the optimal ensemble model is evaluated under varying noise levels. The results show that the prediction error of the optimal model remains within 10%, and R2 is greater than 0.96. Finally, the Shapley Additive Explanations (SHAP) method is applied to perform interpretability analysis on the optimal ROP prediction model, examining the impact of different input factors on the model's predictive performance. Compared to single models and other ensemble models, the proposed ensemble model not only achieves higher accuracy but also demonstrates strong robustness and generalization capability.

Two-step machine learning-aided two-stage hydrothermal liquefaction of biomass for bio-oil upgrading to lower nitrogen content: Experimental verification and parameter optimization

The two-stage hydrothermal liquefaction (HTL) is a promising process for converting biomass into low-nitrogen bio-oil. However, for two-stage HTL, there is a considerable lack of application of artificial intelligence tool. In this study, a two-step machine learning (ML) model was developed to optimizing bio-oil production of two-stage HTL. A total of 644 data points was collected, and a novel feature NH2-C (the mass ratio of amino to carbon) was used to enhancing ML models using Random Forest (RF), Gradient Boosting Regression (GBR), and Extreme Gradient Boosting (XGB) algorithms. The XGB multi-task model demonstrated the best performance (train R2 of 0.98, test R2 of 0.94 and 0.83). The nitrogen content in feedstock is the decisive factor for total nitrogen and nitrogen recovery rate of aqueous phase in HTL stage Ⅰ, while NH2-C for yield and nitrogen content of bio-oil in stage Ⅱ. The two-step ML model performed excellently in 22 validation experiments (with all errors ≤13.68%), and guided the bio-oil preparation with lower nitrogen content (5.46%) from Chlorella. This study provides a new approach for the application of multi-step ML in biomass HTL.

Differential Radiomics-Based Signature Predicts Lung Cancer Risk Accounting for Imaging Parameters in NLSTCohort.

Lung cancer remains the leading cause of cancer-related mortality worldwide, with most cases diagnosed at advanced stages. Hence, there is a need to develop effective predictive models for early detection. This study aims to investigate the impact of imaging parameters and delta radiomic features from temporal scans on lung cancer risk prediction. Using the National Lung Screening Trial (NLST) within a nested case-control study involving 462 positive screenings, radiomic features were extracted from temporal computed tomography (CT) scans and harmonized with ComBat method to adjust variations in slice thickness category (TC) and reconstruction kernel type (KT). Both harmonized and non-harmonized features from baseline (T0), delta features between T0 and a year later (T1), and combined T0 and delta features were utilized for the analysis. Feature reduction was done using LASSO, followed by five feature selection (FS) methods and nine machine learning (ML) models, evaluated with 5-fold cross-validation repeated 10 times. Synthetic Minority Oversampling Technique (SMOTE) was applied to address class imbalances for lung cancer risk prediction. Models using delta features outperformed baseline features, with SMOTE consistently boosting performance when using combination of baseline and delta features. TC-based harmonized features improved performance with SMOTE, but overall, harmonization did not significantly enhance the model performance. The highest test score of 0.76 was achieved in three scenarios: delta features with a Gradient Boosting (GB) model (TC-based harmonization and MultiSurf FS); and T0 + delta features, with both a Support Vector Classifier (SVC) model (KT-based harmonization and F-test FS), and an XGBoost (XGB) model (TC-based harmonization and Mutual Information (MI) FS), all using SMOTE. This study underscores the significance of delta radiomic features and balanced datasets to improve lung cancer prediction. While our findings are based on a subsample of NLST data, they provide a valuable foundation for further exploration. Further research is needed to assess the impact of harmonization on imaging-derived models. Future investigations should explore advanced harmonization techniques and additional imaging parameters to develop robust radiomics-based biomarkers of lung cancer risk.

MACHINE LEARNING APPROACHES FOR DEMAND FORECASTING: THE IMPACT OF CUSTOMER SATISFACTION ON PREDICTION ACCURACY

This study investigates the effectiveness of various machine learning models in predicting product demand based on customer satisfaction data. Four models—Linear Regression, Random Forest, Gradient Boosting, and Support Vector Machine (SVM)—were evaluated using performance metrics, including Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and R² score. The results indicate that Gradient Boosting achieved the highest accuracy, with an MAE of 2.56, MSE of 12.75, RMSE of 3.57, and R² score of 0.82, effectively capturing the complex, non-linear relationships inherent in customer satisfaction factors. Random Forest also demonstrated strong performance, while Linear Regression and SVM showed limitations in handling intricate datasets. These findings underscore the importance of utilizing advanced machine learning techniques for accurate demand forecasting, highlighting the critical role of customer satisfaction data in enhancing predictive capabilities. The insights gained from this research can guide organizations in optimizing inventory management and improving customer satisfaction in a rapidly evolving market.

Machine learning assisted prediction of specific surface area and nitrogen content of biochar based on biomass type and pyrolysis conditions

Predicting and optimizing the physicochemical properties of biochar is crucial for its applications. The characteristics of biomass and pyrolysis conditions are the main factors influencing these properties. However, the numerous components of biomass and the pyrolysis conditions contribute to the substantial challenge in predicting the physicochemical properties, particularly the specific surface area and the nitrogen content of biochar. In this work, machine learning methods including random forest (RF), gradient boosting decision tree (GBDT) and extreme gradient boosting (XGB) (all with R2 exceeding 0.97) were used to predict and analyze specific surface area of biochar (SSA), N content of biochar (N-char), and yield of biochar (Yield-char). Compositions of biomass and pyrolysis conditions were selected as input variables. The partial dependence plot analysis showed the impact way of each influential factor on the target variable and the interactions among these factors in the pyrolysis process. The feature importance of these models indicated that the influencing factors toward predicting three targets (sorted by importance) were specified as follows: pyrolysis temperature, nitrogen content, and fixed carbon for Yield-char; N and ash for N-char; ash and pyrolysis temperature for SSA. This work provided new insights for understanding pyrolysis process of biomass.

Movie Success Prediction Based on Feature and Trailer Comments Using Ensemble+LSTM Model

Predicting the success of a movie is a very important aspect due to the high risks involved in movie production. The challenge lies in the uncertainty within the movie industry and selecting the appropriate machine learning model. We can combine movie features and sentiment analysis from social media using machine learning techniques to achieve movie success prediction. The methods used for predicting based on movie features are Ensemble models (Random Forest + Gradient Boosting). Meanwhile, the methods used for sentiment analysis of trailer comments is LSTM. The evaluation of the models used is based on RMSE and accuracy calculation. The final prediction of success obtains an RMSE of 0,8807 and an accuracy of 91,19%. This represents an improvement from previous research. Further research is recommended to implement the model in the movie industry

Harnessing interpretable and ensemble machine learning techniques for precision fabrication of aligned micro-fibers

Electrospinning is a robust technique for producing micro/nano-scale fibrous structures, influenced by intricate interplays of fluid dynamics, aerodynamics, and electromagnetic forces. Depending on the desired outcome, these fibers can adopt various morphologies, including solid, tubular, concentric, and gradient. Such morphologies are modulated by parameters such as collector configuration, flow rate, voltage, solution properties, and nozzle dimensions. However, the task of modeling and predicting these multifaceted morphologies remains complex. Aligned microfibers with 3D orientation hold promise in tissue engineering, regenerative medicine, and drug delivery, necessitating meticulous control over the fabrication parameters. In our research, we tapped into machine learning (ML) to address these challenges. Classification ML models were designed to predict fibrous patterns—aligned, random, or jet branching—based on determinants like voltage, flow rate, and collector configurations. Notably, the Random Forest (RF) and Support Vector Machine (SVM) models, especially with radial kernel-trick, displayed outstanding predictive capabilities on the test data. Furthermore, regression-based ML was harnessed to discern fiber alignment coherency and inter-fiber distances. Models such as Lasso and Ridge regression elucidated predictive coefficients for these characteristics, while ensemble models, like gradient-boosting (GB) decision trees (DT), showcased prowess in regression scenarios. Key findings spotlighted the significance of parameters like plate gap for alignment coherency and needle-to-collector distance for inter-fiber spacing. As we strive to gain granular control over micro/nano feature morphology in electrospinning, understanding predictor-response dynamics is imperative. Our investigation underscores the essential role of ML in enhancing both qualitative and quantitative precision in fabricating advanced fibrous structures. Moreover, fusing ML with real-time process monitoring offers groundbreaking potential, particularly in Bio-Fabrication, regenerative medicine, and tissue engineering, where high-precision manufacturing remains a top priority.

Discovery of cybervictimization-associated factors among adolescents: Using machine learning and network analysis

The issue of cybervictimization among adolescents is escalating, presenting a significant public concern. Recent research has turned to use a more robust method, machine learning, to explore important predictors for adolescent cybervictimization. The current study tested an extreme gradient boosting (XGBoost) machine learning algorithm to detect cybervictimization-associated factors among adolescents and used network analysis to explore associations between these factors for future targeted interventions. By combining a 6-month longitudinal design, a total of 1181 Chinese adolescents (the average age was 15.78 ± 1.67 years, 55.9% girls) participated in the study. The XGBoost model with satisfactory performance selected the top 10 features from 22 variables associated with cybervictimization by using SHAP value. The network analysis results indicated that maladaptive cognitive emotion regulation strategy is a central node and it has positive correlations with negative self-schema and depression. The XGBoost model and network analysis were useful methods for discovering and understanding cybervictimization-related factors among adolescents. Moreover, these essential factors could offer insights into future interventions for cybervictimization.

An interpretable surrogate model for H2S solubility forecasting in ionic liquids based on machine learning

Here we investigated four different ML-based models, i.e., gaussian process regression (GPR), extreme gradient boosting (i.e., XGBoost), random forest (RF), and support vector machine (SVM), for predicting the solubility of H2S in various ionic liquids (ILs). The dataset was divided into training and testing sets in an 80:20 ratio while the model performance for all models were evaluated using the coefficient of determination (R2), mean absolute error (MAE), and root mean square error (RMSE). Overall, all models effectively predicted H2S solubility, albeit with varying degrees of performance. The GPR provides the best performance, with R2 of 0.9918, MAE of 0.0090, and RMSE of 0.0147. Following this is the XGBoost model with an R2 value of 0.9827, MAE of 0.0155, and RMSE of 0.0213. The RF model displayed slightly lower performance, with an R2 value of 0.9395, MAE of 0.0261, and RMSE of 0.0398 while the lowest performance was demonstrated by the SVM model, which gave an R2 value of 0.9036, MAE of 0.0402, and RMSE of 0.0508. We used SHAP analysis and identified the pressure, temperature, Estate_VSA3, Estate_VSA5, and MinEStateIndex as the top five dominant input features in our model interpretation. In a nutshell, this work presents new insights into the molecular characteristics that affect the solubility of H2S in ILs, paving future research path in this field.

Machine-Learning Application for Predicting Metabolic Dysfunction-Associated Steatotic Liver Disease Using Laboratory and Body Composition Indicators.

Metabolic dysfunction-associated steatotic liver disease (MASLD) represents a significant global health burden without established curative therapies. Early detection and preventive strategies are crucial for effective MASLD management. This study aimed to develop and validate machine-learning (ML) algorithms for accurate MASLD screening in a geographically diverse, large-scale population. Data from the prospective Fasa Cohort Study, initiated in rural Fars province, Iran (March 2014), were employed for this purpose. The required data were collected using blood tests, questionnaires, liver ultrasonography, and physical examinations. A two-step approach identified key predictors from over 100 variables: (1) statistical selection using mean decrease Gini in random forest and (2) incorporation of clinical expertise for alignment with known MASLD risk factors. The hold-out validation approach (with a 70/30 train/validation split) was utilized, along with 5-fold cross-validation on the validation set. Logistic regression, Naïve Bayes, support vector machine, and light gradient-boosting machine (LightGBM) algorithms were compared for model construction with the same input variables based on area under the receiver operating characteristic curve (AUC), sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy. A total of 6,180 adults (52.7% female) were included in the study, categorized into 4816 non-MASLD and 1364 MASLD cases with a mean age (±standard deviation [SD]) of 48.12 (±9.61) and 49.47 (±9.15) years, respectively. Logistic regression outperformed other ML algorithms, achieving an accuracy of 0.88 (95% confidence interval [CI]: 0.86-0.89) and an AUC of 0.92 (95% CI: 0.90-0.93). Among more than 100 variables, the key predictors included waist circumference, body mass index (BMI), hip circumference, wrist circumference, alanine aminotransferase levels, cholesterol, glucose, high-density lipoprotein, and blood pressure. Integration of ML in MASLD management holds significant promise, particularly in resource-limited rural settings. Additionally, the relative importance assigned to each predictor, particularly prominent contributors such as waist circumference and BMI, offers valuable insights into MASLD prevention, diagnosis, and treatment strategies.

Neural decoding and feature selection methods for closed-loop control of avoidance behavior

Objective.Many psychiatric disorders involve excessive avoidant or defensive behavior, such as avoidance in anxiety and trauma disorders or defensive rituals in obsessive-compulsive disorders. Developing algorithms to predict these behaviors from local field potentials (LFPs) could serve as the foundational technology for closed-loop control of such disorders. A significant challenge is identifying the LFP features that encode these defensive behaviors.Approach.We analyzed LFP signals from the infralimbic cortex and basolateral amygdala of rats undergoing tone-shock conditioning and extinction, standard for investigating defensive behaviors. We utilized a comprehensive set of neuro-markers across spectral, temporal, and connectivity domains, employing SHapley Additive exPlanations for feature importance evaluation within Light Gradient-Boosting Machine models. Our goal was to decode three commonly studied avoidance/defensive behaviors: freezing, bar-press suppression, and motion (accelerometry), examining the impact of different features on decoding performance.Main results.Band power and band power ratio between channels emerged as optimal features across sessions. High-gamma (80-150 Hz) power, power ratios, and inter-regional correlations were more informative than other bands that are more classically linked to defensive behaviors. Focusing on highly informative features enhanced performance. Across 4 recording sessions with 16 subjects, we achieved an average coefficient of determination of 0.5357 and 0.3476, and Pearson correlation coefficients of 0.7579 and 0.6092 for accelerometry jerk and bar press rate, respectively. Utilizing only the most informative features revealed differential encoding between accelerometry and bar press rate, with the former primarily through local spectral power and the latter via inter-regional connectivity. Our methodology demonstrated remarkably low training/inference time and memory usage, requiring<310 ms for training,<0.051 ms for inference, and 16.6 kB of memory, using a single core of AMD Ryzen Threadripper PRO 5995WX CPU.Significance.Our results demonstrate the feasibility of accurately decoding defensive behaviors with minimal latency, using LFP features from neural circuits strongly linked to these behaviors. This methodology holds promise for real-time decoding to identify physiological targets in closed-loop psychiatric neuromodulation.

Superphot+: Real-time Fitting and Classification of Supernova Light Curves

Photometric classifications of supernova (SN) light curves have become necessary to utilize the full potential of large samples of observations obtained from wide-field photometric surveys, such as the Zwicky Transient Facility (ZTF) and the Vera C. Rubin Observatory. Here, we present a photometric classifier for SN light curves that does not rely on redshift information and still maintains comparable accuracy to redshift-dependent classifiers. Our new package, Superphot+, uses a parametric model to extract meaningful features from multiband SN light curves. We train a gradient-boosted machine with fit parameters from 6061 ZTF SNe that pass data quality cuts and are spectroscopically classified as one of five classes: SN Ia, SN II, SN Ib/c, SN IIn, and SLSN-I. Without redshift information, our classifier yields a class-averaged F 1-score of 0.61 ± 0.02 and a total accuracy of 0.83 ± 0.01. Including redshift information improves these metrics to 0.71 ± 0.02 and 0.88 ± 0.01, respectively. We assign new class probabilities to 3558 ZTF transients that show SN-like characteristics (based on the ALeRCE Broker light-curve and stamp classifiers) but lack spectroscopic classifications. Finally, we compare our predicted SN labels with those generated by the ALeRCE light-curve classifier, finding that the two classifiers agree on photometric labels for 82% ± 2% of light curves with spectroscopic labels and 72% ± 0% of light curves without spectroscopic labels. Superphot+ is currently classifying ZTF SNe in real time via the ANTARES Broker, and is designed for simple adaptation to six-band Rubin light curves in the future.

A robust LightGBM model for concrete tensile strength forecast to aid in resilience-based structure strategies

The tensile strength (TS) of concrete is a factor in the design of civil engineering structures. Traditional methods for assessing concrete TS have notable limitations, leading to the emergence of non-destructive testing (NDT) as an innovative alternative. This study introduces a novel, optimized approach for accurate, non-destructive TS prediction by integrating Gradio, Bayesian optimization (BO), and Shapley additive explanations (SHAP). The input parameters for the model were ultrasonic pulse velocity (UPV) and electrical resistivity (ER). Using a comprehensive dataset of 3460 datapoints, a light gradient-boosting machine (LightGBM) model was developed. Through extensive hyperparameter tuning facilitated by BO, the model achieved a robust prediction accuracy of approximately 98 %. The novel application of SHAP in this study provided deep insights into model behavior, confirming UPV as the most critical variable for TS prediction. The deployment of the model using Gradio further enhances its accessibility and usability. This combined methodology not only advances the application of machine learning (ML) in civil engineering but also offers a powerful, globally applicable tool for accurately forecasting concrete TS. This supports resilience-based management strategies, essential for the sustainable development of infrastructure. The study's novel integration of BO for optimization and SHAP for interpretability marks a significant step forward in the precise prediction of TS, offering practical benefits for both researchers and industry professionals.

Acoustic emission and machine learning algorithms for particle size analysis in gas-solid fluidized bed reactors

In this work, a combination of an acoustic emission (AE) technique and a machine learning (ML) algorithm (Random Forest (RF) and Gradient Boosting Regressor (GBR)) is developed to characterize the particle size distribution in gas-solid fluidized bed reactors. A theoretical approach to explain the generation of acoustic emission signal in gas-solid flows is presented. An AE signal is generated in gas-solid fluidized beds due to the collision and friction between fluidized particles as well as between particles and the bed inner wall. The generated AE signal is in the form of an elastic wave with frequencies >100 KHz and it propagates through the gas-solid mixture. An inversion algorithm is used to extract the information about the particle size starting from the energy of the AE signal. The advantages of this AE technique are that it is a cheap, sensitive, non-intrusive, radiation-free, suitable for on-line measurements. Combining this AE technique with ML algorithms is beneficial for applications to industrial settings, reducing the cost of signal post-processing. Experiments were conducted in a pseudo-2D flat fluidized bed with four glass bead samples, with sizes ranging from 100 μm to 710 μm. AE signals were recorded with a sampling frequency of 5 MHz. The AE signal post-processing and data preparation for the ML process are explained. For the ML process, the AE frequency, AE energy and particle collision velocity data sets were divided into training (60%), cross-validation (20%) and test sets (20%). Two ensemble ML approaches, namely Random Forest and Gradient Boosting Regressor, are applied to predict particle sizes based on the AE signal features. The combination of these two models results in a coefficient of determination (R2) value greater than 0.9504.

A Survey of AI Methods for Detection of DDoS Attacks on Networks

This survey explores various artificial intelligence (AI) methods for detecting Distributed Denial of Service (DDoS) attacks on networks. It classifies these approaches into machine learning, deep learning, and other AI-based techniques, providing a comprehensive overview of current advancements in the field. Numerous research studies in the field of machine learning have evaluated DDoS attack detection performance using various datasets and techniques. Some noteworthy results are the supremacy of the J48 algorithm in SDN networks and the efficacy of the AdaBoost and Gradient Boost classifiers. In other investigations, Random Forest, Support Vector Machine, and Naive Bayes also showed excellent accuracy rates, up to 99.7%. To improve DDoS detection, deep learning techniques introduced autoencoders, hybrid models, and recurrent neural networks. These models achieved accuracy rates as high as 99.99%, frequently outperforming more conventional machine learning techniques. Enhanced detection rates were achieved by the utilization of a varied dataset in conjunction with deep-stacked autoencoders. Artificial intelligence methods such as Fuzzy Logic, Artificial Bee Colony, Ant Colony Optimization, and Whale Optimization Algorithm were used to identify DDoS assaults. These methods demonstrated high accuracy rates, efficient detection of various attack types, and improvements in reducing false positives; the integration of these techniques into intrusion detection systems offers a strong defense against dynamic DDoS threats. The overall survey highlights the effectiveness of AI techniques in DDoS attack detection across various methodologies.

A novel FDEM-GSA method with applications in deformation and damage analysis of surrounding rock in deep-buried tunnels

In underground hydraulic tunnel engineering, particularlydeep-buried projects, the deformations and stability of the surrounding rock are critical to the construction process. Due to the complex depositional environment of such tunnels, multiple factors influence the stability of the surrounding rock during construction. This paper proposes a novel hybrid method combining Finite-Discrete Element Method (FDEM) with Global Sensitivity Analysis (GSA) and machine learning. The FDEM model is used to establish an approximate relationship between input and output parameters, and its accuracy is validated through comparison with monitoring data. On the basis of data simulated by the FDEM model, a meta-model is developed by Particle Swarm Optimization-Extreme Gradient Boosting (PSO-XGBoost). 10,000 data sets are generated using the meta-model for Sobol sensitivity analysis. The proposed analysis framework is applied to study the deep-buried water diversion tunnel in the Central Yunnan Water Diversion Project (CYWDP). The results indicate that in deep-buried tunnel environments, as the tunnel depth increases, the dominant factor influencing the deformation of the surrounding rock transitions from tunnel depth to the mechanical properties of the rock itself. These findings provide valuable insights for optimizing construction plans in similar tunnel projects and offer a new perspective on sensitivity analysis frameworks based on numerical computations.

APPLICATION OF MACHINE LEARNING METHODS TO ENHANCE THE PERFORMANCE OF BIG DATA SORTING ALGORITHMS

The use of machine learning methods to optimize big data sorting algorithms has become an urgent research topic due to the growing volume of information and the requirements for their rapid processing. Machine learning provides opportunities to automate and improve traditional sorting methods, allowing you to reduce the cost of computing resources and time. This is achieved by analyzing the characteristics of the data and preprocessing them using classification and regression. The main advantages of using machine learning in sorting big data include improving the accuracy and adaptability of algorithms to different types of data, which is especially important for areas with large amounts of information, such as finance, medicine and logistics. Progressive machine learning algorithms such as supporting vectors, decision trees, and gradient boosting demonstrate high efficiency and potential for further development and integration.

Using Machine Learning and Electronic Health Records to Identify Neuropsychiatric Risk Scores for Delirium in ICU and General Hospital Settings.

Delirium is a common and acute neuropsychiatric syndrome that requires timely intervention to prevent its associated morbidity and mortality. Yet, its diagnosis and symptoms are often overlooked due to its variable clinical presentation and fluctuating nature. Thus, in this study, we address the barriers to delirium diagnosis by utilizing a machine learning-based predictive algorithm for incident delirium that relies on archived electronic health records (EHRs) data. We used the Medical Information Mart for Intensive Care (MIMIC) database to create a detailed dataset for identifying delirium in intensive care unit (ICU) patients. Our approach involved training machine learning models on this dataset to pinpoint critical clinical features for delirium detection. These features were then refined and applied to non-ICU patients using EHRs from the American University of Beirut Medical Center (AUBMC). Our study assessed machine learning models like Extreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), Classification and Regression Trees (CART), Random Forest (RF), Neural Oblivious Decision Ensembles (NODE), and Logistic Regression (LR), highlighting superior delirium detection in diverse clinical settings. The CatBoost model excelled in ICU environments with an F1 Score of 89.2%, while XGBoost performed best in general hospital settings with a 75.4% F1 Score. Interpretations using Tabular Local Interpretable Model-agnostic Explanations (LIME) revealed critical indicators such as prothrombin time and hematocrit levels, enhancing model transparency and clinical applicability. These clinical insights help differentiate the delirium predictors between ICU patients, who are often sensitive to various factors. The proposed predictive algorithm improves delirium detection rates and streamlines efficiency in hospital electronic systems, thereby enabling prompt interventions to prevent delirium progression and associated complications. The clinical indicators for delirium that we identified in general hospital settings and ICU can greatly help healthcare professionals identify potential causes of delirium and reduce misdiagnosis.

Development and internal validation of prehospital prediction models for identifying intracerebral haemorrhage in suspected stroke patients

IntroductionDistinguishing patients with intracerebral haemorrhage (ICH) from other suspected stroke cases in the prehospital setting is crucial for determining the appropriate level of care and minimising the onset-to-treatment time, thereby...