Assess Model Performance Research Articles

SentinelFusion based machine learning comprehensive approach for enhanced computer forensics.

In the rapidly evolving landscape of modern technology, the convergence of blockchain innovation and machine learning advancements presents unparalleled opportunities to enhance computer forensics. This study introduces SentinelFusion, an ensemble-based machine learning framework designed to bolster secrecy, privacy, and data integrity within blockchain systems. By integrating cutting-edge blockchain security properties with the predictive capabilities of machine learning, SentinelFusion aims to improve the detection and prevention of security breaches and data tampering. Utilizing a comprehensive blockchain-based dataset of various criminal activities, the framework leverages multiple machine learning models, including support vector machines, K-nearest neighbors, naive Bayes, logistic regression, and decision trees, alongside the novel SentinelFusion ensemble model. Extensive evaluation metrics such as accuracy, precision, recall, and F1 score are used to assess model performance. The results demonstrate that SentinelFusion outperforms individual models, achieving an accuracy, precision, recall, and F1 score of 0.99. This study's findings underscore the potential of combining blockchain technology and machine learning to advance computer forensics, providing valuable insights for practitioners and researchers in the field.

Enhancing solar power forecasting with machine learning using principal component analysis and diverse statistical indicators

_Predicting solar energy is essential for efficient power system planning and the successful integration of renewable energy sources. This study aims to develop a framework for evaluating various machine learning models and feature selection strategies for solar energy prediction. The research applies six machine learning models, i.e., linear regression (LR), random forest (RF), neural networks (NN), K-nearest neighbor (KNN), gradient boosting (GB), and AdaBoost, to datasets from 2019 to 2021 collected at the Abiod Sid Cheikh solar station in southern Algeria. Various statistical indicators, including R2, RMSE, MAE, and Adj-R, were analyzed to assess model performance. The analysis revealed that R2 values ranged from 0.591 to 0.996 kW/m2, RMSE from 0.510 to 1.78 kW/m2, and MAE from 0.357 to 0.856 kW/m2 across different models. KNN and NN models showed significant errors, while GB and RF models demonstrated strong accuracies (RMS = 0.9). AdaBoost and LR excelled in real-time and short-term predictions, exhibiting an RMS of 0.99. This framework offers a comprehensive evaluation method for selecting the most suitable machine learning models and feature selection strategies for solar energy prediction. The findings can assist energy planners and engineers in choosing the appropriate models for accurate solar energy prediction, thereby enhancing the efficiency of power systems. Improved solar energy prediction can contribute to more reliable integration of renewable energy into power grids, supporting the transition to cleaner energy sources and reducing environmental impacts.

Confronting transitions in fishery fleet structure and selectivity: practical recommendations for integrated age-structured stock assessments based on simulation analysis

Dynamic shifts in fleet structure and gear usage lead to complex implications for representing fishery selectivity in stock assessment models. There is generally a lack of consensus on how assessment models should be configured to confront changes in fishery fleet structure or associated selectivity forms, while balancing complexity-parsimony tradeoffs. We conducted a simulation analysis to evaluate the performance of alternative assessment models when confronted with fleet transitions among gear types, which included differences in (1) rates of transition (i.e., a fast or slow transition among gears), and (2) selectivity forms for each modeled fleet (i.e., asymptotic or dome-shaped). In general, explicitly modeling fleet structure (i.e., multi-fleet models) performed well, but demonstrated bias in biomass estimates and management reference points when selectivity forms were mis-specified. Single-fleet models were only unbiased when time-varying selectivity (e.g., using time blocks or continuous formulations) was estimated to account for changes among gear types. Our results suggest that single-fleet models with time-varying fishery selectivity are adequate for operational management advice, but research oriented multi-fleet models should be used as validation tools to explore model consistency within single-fleet models.

Machine Learning Models for Predicting Risky Pregnancies in Early Clinical Interventions

Background: Risky pregnancies present significant challenges in maternal healthcare, often requiring accurate prediction to prevent adverse outcomes. Machine learning (ML) models offer a promising approach for predicting such risks, enabling timely interventions. This study evaluates five machine learning models—Logistic Regression, Decision Tree, K-Nearest Neighbors (KNN), Naive Bayes, and Support Vector Machine (SVM)—for their effectiveness in predicting risky pregnancies using clinical datasets. Methods: The study developed and evaluated five ML models, each implemented using Python’s scikit-learn library. The dataset was split into 75% for training and 25% for testing. Standard classification metrics, including accuracy, precision, recall, and F1-score, were used to assess model performance. Hyperparameter tuning was conducted using grid search and cross-validation to optimize model parameters. The models’ performance was compared to identify the most suitable for clinical applications. Results: The Decision Tree model achieved the highest accuracy (100% on training data, 95.6% on testing data), along with excellent precision, recall, and F1-scores for both classes, making it the most accurate and interpretable model for predicting risky pregnancies. Logistic Regression also performed well, particularly in identifying high-risk cases, with testing accuracy of 82%. KNN and SVM provided moderate accuracy, with KNN achieving 78% testing accuracy and SVM 80%. Naive Bayes, however, performed poorly, achieving only 43.2% accuracy due to its assumption of feature independence, which was not suitable for the dataset. Conclusion: The Decision Tree and Logistic Regression models emerged as the most effective for predicting risky pregnancies, offering high accuracy and interpretability, crucial for clinical decision-making.

Exploring Machine Learning Techniques for Open Stope Stability Prediction: A Comparative Study and Feature Importance Analysis

The stability of underground excavations is essential for ensuring the safety of mining operations. Classical stability assessment methods, established in empirical formulas and rock mass classification systems, have long been employed for evaluating stope stability in underground mining. Stability graphs, a popular empirical approach, utilize factors like rock stress, joint orientation, and surface orientation to calculate stability numbers critical for stope design. However, modern advancements in machine learning present new opportunities for enhancing predictive capabilities and understanding complex relationships influencing stope stability. Building upon research demonstrating the feasibility of using machine learning for stability prediction, our study investigates and compares several machine learning algorithms. By analyzing a dataset comprising stope dimensions and geomechanical properties, we explore the potential of machine learning models such as Random Forest, Support Vector Machine, AdaBoost, XGBoost, LightGBM, and Artificial Neural Network in predicting stope stability. Evaluation metrics including accuracy, precision, recall, and F1 score are employed to assess model performance, with the Artificial Neural Network emerging as the most effective. Furthermore, SHapley Additive exPlanations (SHAP) analysis enhances interpretability by explaining the contribution of individual features to model predictions.

Implementation of Convolutional Neural Network Algorithms to Detect the Ripeness of Palm Fruits based on Image Colors

The ripeness of oil palm fruit greatly affects the quality of the oil produced. Usually, humans manually detect the maturity of oil palm fruit, a process that can be inconsistent and time-consuming. Therefore, there is a need for automated methods that can improve accuracy and efficiency in determining the maturity of oil palm fruit. The study proposes using the Convolutional Neural Network algorithm to detect oil palm fruit maturity based on digital color images. We expect a convolutional neural network, known for its effectiveness in image analysis, to deliver accurate results in classifying fruit maturity. The aim of this study is to develop and test a model of a convolutional neural network that can classify oil palm fruit into three maturity categories: raw, ripe, and rotten. This study employs a maturity-categorized data set of oil palm images. We built and trained a convolutional neural network model using this data set. We evaluate the model performance using four main metrics: accuracy, precision, recall, and f1-score. We perform an in-depth analysis to assess model performance in each maturity category. The confusion matrix yielded an accuracy of 74.6%. The convolutional neural network model developed showed the highest precision results for the ripe fruit (86.00%), followed by rotten fruit (72.30%), and raw fruit (67.30%). We obtained the highest recalls (85.40%) for the raw fruit category, followed by ripe fruits (75.50%) and rotten fruits (72.30%.

Recent advances in crack detection technologies for structures: a survey of 2022-2023 literature

IntroductionCracks, as structural defects or fractures in materials like concrete, asphalt, and metal, pose significant challenges to the stability and safety of various structures. Addressing crack detection is of paramount importance due to its implications for public safety, infrastructure integrity, maintenance costs, asset longevity, preventive maintenance, economic impact, and environmental considerations.MethodsIn this survey paper, we present a comprehensive analysis of recent advancements and developments in crack detection technologies for structures, with a specific focus on articles published between 2022 and 2023. Our methodology involves an exhaustive search of the Scopus database using keywords related to crack detection and machine learning techniques. Among the 129 papers reviewed, 85 were closely aligned with our research focus.ResultsWe explore datasets that underpin crack detection research, categorizing them as public datasets, papers with their own datasets, and those using a hybrid approach. The prevalence and usage patterns of public datasets are presented, highlighting datasets like Crack500, Crack Forest Dataset (CFD), and Deep Crack. Furthermore, papers employing proprietary datasets and those combining public and proprietary sources are examined. The survey comprehensively investigates the algorithms and methods utilized, encompassing CNN, YOLO, UNet, ResNet, and others, elucidating their contributions to crack detection. Evaluation metrics such as accuracy, precision, recall, F1-score, and IoU are discussed in the context of assessing model performance. The results of the 85 papers are summarized, demonstrating advancements in crack detection accuracy, efficiency, and applicability.DiscussionNotably, we observe a trend towards using modern and novel algorithms, such as Vision Transformers (ViT), and a shift away from traditional methods. The conclusion encapsulates the current state of crack detection research, highlighting the integration of multiple algorithms, expert models, and innovative data collection techniques. As a future direction, the adoption of emerging algorithms like ViT is suggested. This survey paper serves as a valuable resource for researchers, practitioners, and engineers working in the field of crack detection, offering insights into the latest trends, methodologies, and challenges.

Pre-operative prediction of histopathological growth patterns of colorectal cancer liver metastasis using MRI-based radiomic models.

Histopathological growth patterns (HGPs) of colorectal liver metastases (CRLMs) have prognostic value. However, the differentiation of HGPs relies on postoperative pathology. This study aimed to develop a magnetic resonance imaging (MRI)-based radiomic model to predict HGP pre-operatively, following the latest guidelines. This retrospective study included 93 chemotherapy-naïve patients with CRLMs who underwent contrast-enhanced liver MRI and a partial hepatectomy between 2014 and 2022. Radiomic features were extracted from the tumor zone (RTumor), a 2-mm outer ring (RT+2), a 2-mm inner ring (RT-2), and a combined ring (R2+2) on late arterial phase MRI images. Analysis of variance method (ANOVA) and least absolute shrinkage and selection operator (LASSO) algorithms were used for feature selection. Logistic regression with five-fold cross-validation was used for model construction. Receiver operating characteristic curves, calibrated curves, and decision curve analyses were used to assess model performance. DeLong tests were used to compare different models. Twenty-nine desmoplastic and sixty-four non-desmoplastic CRLMs were included. The radiomic models achieved area under the curve (AUC) values of 0.736, 0.906, 0.804, and 0.794 for RTumor, RT-2, RT+2, and R2+2, respectively, in the training cohorts. The AUC values were 0.713, 0.876, 0.785, and 0.777 for RTumor, RT-2, RT+2, and R2+2, respectively, in the validation cohort. RT-2 exhibited the best performance. The MRI-based radiomic models could predict HGPs in CRLMs pre-operatively.

The importance of biotic interactions in distribution models of wild bees depends on the type of ecological relations, spatial scale and range

Studies have found that biotic information can play an important role in shaping the distribution of species even at large scales. However, results from species distribution models are not always consistent among studies and the underlying factors that influence the importance of biotic information to distribution models, are unclear. We studied wild bees and plants, and cleptoparasite bees and their hosts in the Netherlands to evaluate how the inclusion of their biotic interactions affects the performance of species distribution models. We assessed model performance through spatial block cross‐validation and by comparing models with interactions to models where the interacting species was randomized. Finally, we evaluated how, 1) spatial resolution, 2) taxonomic rank (genus or species), 3) degree of specialization, 4) distribution of the biotic factor, 5) bee body size and 6) type of biotic interaction, affect the importance of biotic interactions in shaping the distribution of wild bee species using generalized linear models (GLMs). We found that the models of wild bees improved when the biotic factor was included. The model performance improved the most for parasitic bees. Spatial resolution, taxonomic rank, distribution range of the biotic factor and degree of specialization of the modelled species all influenced the importance of the biotic interaction to the models. We encourage researchers to include biotic interactions in species distribution models, especially for specialized species and when the biotic factor has a limited distribution range. However, before adding the biotic factor we suggest considering different spatial resolutions and taxonomic ranks of the biotic factor. We recommend using single species or genus data as a biotic factor in the models of specialist species and for the generalist species, we recommend using an approximate measure of interactions, such as flower richness.

Optimizing fraud detection in financial transactions with machine learning and imbalance mitigation

AbstractThe rapid advancement of the Internet and digital payments has transformed the landscape of financial transactions, leading to both technological progress and an alarming rise in cybercrime. This study addresses the critical issue of financial fraud detection in the era of digital payments, focusing on enhancing operational risk frameworks to mitigate the increasing threats. The objective is to improve the predictive performance of fraud detection systems using machine learning techniques. The methodology involves a comprehensive data preprocessing and model creation process, including one‐hot encoding, feature selection, sampling, standardization, and tokenization. Six machine learning models are employed for fraud detection, and their hyperparameters are optimized. Evaluation metrics such as accuracy, precision, recall, and F1‐score are used to assess model performance. Results reveal that XGBoost and Random Forest outperform other models, achieving a balance between false positives and false negatives. The study meets the requirements for fraud detection systems, ensuring accuracy, scalability, adaptability, and explainability. This paper provides valuable insights into the efficacy of machine learning models for financial fraud detection and emphasizes the importance of striking a balance between false positives and false negatives.

Enhancing genomic prediction with Stacking Ensemble Learning in Arabica Coffee.

Coffee Breeding programs have traditionally relied on observing plant characteristics over years, a slow and costly process. Genomic selection (GS) offers a DNA-based alternative for faster selection of superior cultivars. Stacking Ensemble Learning (SEL) combines multiple models for potentially even more accurate selection. This study explores SEL potential in coffee breeding, aiming to improve prediction accuracy for important traits [yield (YL), total number of the fruits (NF), leaf miner infestation (LM), and cercosporiosis incidence (Cer)] in Coffea Arabica. We analyzed data from 195 individuals genotyped for 21,211 single-nucleotide polymorphism (SNP) markers. To comprehensively assess model performance, we employed a cross-validation (CV) scheme. Genomic Best Linear Unbiased Prediction (GBLUP), multivariate adaptive regression splines (MARS), Quantile Random Forest (QRF), and Random Forest (RF) served as base learners. For the meta-learner within the SEL framework, various options were explored, including Ridge Regression, RF, GBLUP, and Single Average. The SEL method was able to predict the predictive ability (PA) of important traits in Coffea Arabica. SEL presented higher PA compared with those obtained for all base learner methods. The gains in PA in relation to GBLUP were 87.44% (the ratio between the PA obtained from best Stacking model and the GBLUP), 37.83%, 199.82%, and 14.59% for YL, NF, LM and Cer, respectively. Overall, SEL presents a promising approach for GS. By combining predictions from multiple models, SEL can potentially enhance the PA of GS for complex traits.

Predictive Modeling of Long-Term Prognosis After Resection in Typical Pulmonary Carcinoid: A Machine Learning Perspective

Typical Pulmonary Carcinoid (TPC) is defined by its slow growth, frequently necessitating surgical intervention. Despite this, the long-term outcomes following tumor resection are not well understood. This study examined the factors impacting Overall Survival (OS) in patients with TPC, leveraging data from the Surveillance, Epidemiology, and End Results database spanning from 2000 to 2018. We employed Lasso-Cox analysis to identify prognostic features and developed various models using Random Forest, XGBoost, and Cox regression algorithms. Subsequently, we assessed model performance using metrics such as Area Under the Curve (AUC), calibration plot, Brier score, and Decision Curve Analysis (DCA). Among the 2687 patients, we identified five clinical features significantly affecting OS. Notably, the Random Forest model exhibited strong performance, achieving 5- and 7-year AUC values of 0.744/0.757 in the training set and 0.715/0.740 in the validation set, respectively, outperforming other models. Additionally, we developed a web-based platform aimed at facilitating easy access to the model. This study presents a machine learning model and a web-based support system for healthcare professionals, assisting in personalized treatment decisions for patients with TPC post-tumor resection.

Evaluating and Updating the IMPACT Model to Predict Outcomes in Two Contemporary North American Traumatic Brain Injury Cohorts.

The International Mission on Prognosis and Analysis of Clinical Trials in Traumatic Brain Injury (IMPACT) model is a widely recognized prognostic model applied after traumatic brain injury (TBI). However, it was developed with patient cohorts that may not reflect modern practice patterns in North America. We analyzed data from two sources: the placebo arm of the phase II double-blinded, multicenter, randomized controlled trial Prehospital Tranexamic Acid for TBI (TXA) cohort and an observational cohort with similar inclusion/exclusion criteria (Predictors of Low-risk Phenotypes after Traumatic Brain Injury Incorporating Proteomic Biomarker Signatures [PROTIPS] cohort). All three versions of the IMPACT model-core, extended, and laboratory-were evaluated for 6-month mortality (Glasgow Outcome Scale Extended [GOSE] = 1) and unfavorable outcomes (GOSE = 1-4). Calibration (intercept and slope) and discrimination (area under the receiver operating characteristic curve [ROC-AUC]) were used to assess model performance. We then compared three model updating methods-recalibration in the large, logistic recalibration, and coefficient update-with the best update method determined by likelihood ratio tests. In our calibration analysis, recalibration improved both intercepts and slopes, indicating more accurate predicted probabilities when recalibration was done. Discriminative performance of the IMPACT models, measured by AUC, showed mortality prediction ROCs between 0.61 and 0.82 for the TXA cohort, with the coefficient updated Lab model achieving the highest at 0.84. Unfavorable outcomes had lower AUCs, ranging from 0.60 to 0.79. Similarly, in the PROTIPS cohort, AUCs for mortality ranged from 0.75 to 0.82, with the coefficient updated Lab model also showing superior performance (AUC 0.84). Unfavorable outcomes in this cohort presented AUCs from 0.67 to 0.73, consistently lower than mortality predictions. The closed testing procedure using likelihood ratio tests consistently identified the coefficient update model as superior, outperforming the original and recalibrated models across all cohorts. In our comprehensive evaluation of the IMPACT model, the coefficient updated models were the best performing across all cohorts through a structured closed testing procedure. Thus, standardization of model updating procedures is needed to reproducibly determine the best performing versions of IMPACT that reflect the specific characteristics of a dataset.

Identifying Dementia Severity Among People Living With Dementia Using Administrative Claims Data

ObjectivesThere is currently no reliable tool for classifying dementia severity level based on administrative claims data. We aimed to develop a claims-based model to identify patients with severe dementia among a cohort of patients with dementia. DesignRetrospective cohort study. Setting and ParticipantsWe identified people living with dementia (PLWD) in US Medicare claims data linked with the Minimum Data Set (MDS) and Outcome and Assessment Information Set (OASIS). MethodsSevere dementia was defined based on cognitive and functional status data available in the MDS and OASIS. The dataset was randomly divided into training (70%) and validation (30%) sets, and a logistic regression model was developed to predict severe dementia using baseline (assessed in the prior year) features selected by generalized linear mixed models (GLMMs) with least absolute shrinkage and selection operator (LASSO) regression. We assessed model performance by area under the receiver operating characteristic curve (AUROC), area under precision-recall curve (AUPRC), and precision and recall at various cutoff points, including Youden Index. We compared the model performance with and without using Synthetic Minority Oversampling Technique (SMOTE) to reduce the imbalance of the dataset. ResultsOur study cohort included 254,410 PLWD with 17,907 (7.0%) classified as having severe dementia. The AUROC of our primary model, without SMOTE, was 0.81 in the training and 0.80 in the validation set. In the validation set at the optimized Youden Index, the model had a sensitivity of 0.77 and specificity of 0.70. Using a SMOTE-balanced validation set, the model had an AUROC of 0.83, AUPRC of 0.80, sensitivity of 0.79, specificity of 0.74, positive predictive value of 0.75, and negative predictive value of 0.78 when at the optimized Youden Index. Conclusions and ImplicationsOur claims-based algorithm to identify patients living with severe dementia can be useful for claims-based pharmacoepidemiologic and health services research.

Diagnostics for categorical response models based on quantile residuals and distance measures

Polytomous categorical data are frequent in studies, that can be obtained with an individual or grouped structure. In both structures, the generalized logit model is commonly used to relate the covariates on the response variable. After fitting a model, one of the challenges is the definition of an appropriate residual and choosing diagnostic techniques. Since the polytomous variable is multivariate, raw, Pearson, or deviance residuals are vectors and their asymptotic distribution is generally unknown, which leads to difficulties in graphical visualization and interpretation. Therefore, the definition of appropriate residuals and the choice of the correct analysis in diagnostic tools is important, especially for nominal data, where a restriction of methods is observed. This paper proposes the use of randomized quantile residuals associated with individual and grouped nominal data, as well as Euclidean and Mahalanobis distance measures, as an alternative to reduce the dimension of the residuals. We developed simulation studies with both data structures associated. The half-normal plots with simulation envelopes were used to assess model performance. These studies demonstrated a good performance of the quantile residuals, and the distance measurements allowed a better interpretation of the graphical techniques. We illustrate the proposed procedures with two applications to real data.

Levodopa-induced dyskinesia in Parkinson's disease: Insights from cross-cohort prognostic analysis using machine learning

BackgroundProlonged levodopa treatment in Parkinson's disease (PD) often leads to motor complications, including levodopa-induced dyskinesia (LID). Despite continuous levodopa treatment, some patients do not develop LID symptoms, even in later stages of the disease. ObjectiveThis study explores machine learning (ML) methods using baseline clinical characteristics to predict the development of LID in PD patients over four years, across multiple cohorts. MethodsUsing interpretable ML approaches, we analyzed clinical data from three independent longitudinal PD cohorts (LuxPARK, n = 356; PPMI, n = 484; ICEBERG, n = 113) to develop cross-cohort prognostic models and identify potential predictors for the development of LID. We examined cohort-specific and shared predictive factors, assessing model performance and stability through cross-validation analyses. ResultsConsistent cross-validation results for single and multiple cohort analyses highlighted the effectiveness of the ML models and identified baseline clinical characteristics with significant predictive value for the LID prognosis in PD. Predictors positively correlated with LID include axial symptoms, freezing of gait, and rigidity in the lower extremities. Conversely, the risk of developing LID was inversely associated with the occurrence of resting tremors, higher body weight, later onset of PD, and visuospatial abilities. ConclusionsThis study presents interpretable ML models for dyskinesia prognosis with significant predictive power in cross-cohort analyses. The models may pave the way for proactive interventions against dyskinesia in PD by optimizing levodopa dosing regimens and adjunct treatments with dopamine agonists or MAO-B inhibitors, and by employing non-pharmacological interventions such as dietary adjustments affecting levodopa absorption for high-risk LID patients.

P-757 An artificial intelligence model to predict upcoming patient oocyte collections to help optimize efficiency and safety in the embryology laboratory

Abstract Study question Can artificial intelligence (AI) be used to forecast the number of oocyte collections that will occur at a clinic multiple days in advance? Summary answer An AI model can accurately predict how many oocyte collections will occur across a clinic up to seven days in advance. What is known already There exists significant variability in the duration of ovarian stimulation for patients going through IVF, which can often lead to peaks and valleys of workload for the embryology laboratory. On a given day, a laboratory being under or over resourced can lead to poor utilization of resources or overtime hours. This is especially problematic as the demand for IVF steadily increases, particularly for large clinics doing thousands of oocyte collections (OC) per year. A tool to help predict clinic OC workload many days in advance could assist with laboratory staffing and operational efficiency through load leveling. Study design, size, duration Historical, de-identified electronic medical record data from 6140 patients undergoing ovarian stimulation at a large IVF clinic in the U.S. between January 2021 to January 2023 was collected. Patients who started treatment between January 2021 to December 2021 were used for training the AI model (N = 3126). The remaining patients were used for evaluating the model (N = 3014). Participants/materials, setting, methods A gradient boosting classifier (Catboost) was trained to predict a patient’s remaining cycle duration using E2, follicle sizes, and day of medication. This model outputs a probability that a patient’s OC will land on any specific future day. To predict the number of OC across a clinic on a future day, the model predicts the probability that each patient will have their retrieval on that day, and calculates the convolution to estimate total retrievals. Main results and the role of chance The clinic had an average of 7.9 OC per day. The OC projection accuracy was assessed by iterating through each date in the test dataset and predicting the number of OC each day over the next 7 days, for all patients undergoing stimulation who had not yet been triggered. In this analysis, the model predicted the total number of OC on each day with a mean absolute error of 1.56 retrievals and an R2 of 0.72. The model was able to identify days where workload exceeded 11 OC (80th percentile) with 86% accuracy and 58% precision. The model was able to identify days where workload was below 5 OC (20th percentile) with 95% accuracy and 91% precision. Limitations, reasons for caution This model was trained at a single clinic and evaluated retrospectively. Future work should assess model performance at clinics with varying patient throughput, and evaluate prospectively how this tool could assist with load leveling of OC volume. Wider implications of the findings An AI model can predict when patients will be triggered based on their medication and response to treatment. This tool can be used to forecast the total number of OC at a clinic up to 7 days in advance, which could help improve efficiency for embryology and clinical staff. Trial registration number not applicable

Development and Validation of the Hospital Medicine Safety Sepsis Initiative Mortality Model

When comparing outcomes after sepsis, it is essential to account for patient case mix to make fair comparisons. We developed a model to assess risk-adjusted 30-day mortality in the Michigan Hospital Medicine Safety sepsis initiative (HMS-Sepsis). Can HMS-Sepsis registry data adequately predict risk of 30-day mortality? Do performance assessments using adjusted vsunadjusted data differ? Retrospective cohort of community-onset sepsis hospitalizations in the HMS-Sepsis registry (April 2022-September 2023), with split-derivation (70%) and validation (30%) cohorts. We fit a risk-adjustment model (HMS-Sepsis mortality model) incorporating acute physiologic, demographic, and baseline health data and assessed model performance using concordance (C) statistics, Brier's scores, and comparisons of predicted vsobserved mortality by deciles of risk. We compared hospital performance (first quintile, middle quintiles, fifth quintile) using observed vsadjusted mortality to understand the extent to which risk adjustment impacted hospital performance assessment. Among 17,514 hospitalizations from 66 hospitals during the study period, 12,260 hospitalizations (70%) were used for model derivation and 5,254 hospitalizations (30%) were used for model validation. Thirty-day mortality for the total cohort was 19.4%. The final model included 13 physiologic variables, two physiologic interactions, and 16 demographic and chronic health variables. The most significant variables were age, metastatic solid tumor, temperature, altered mental status, and platelet count. The model C statistic was 0.82 for the derivation cohort, 0.81 for the validation cohort, and≥ 0.78 for all subgroups assessed. Overall calibration error was 0.0%, and mean calibration error across deciles of risk was 1.5%. Standardized mortality ratios yielded different assessments than observed mortality for 33.9%of hospitals. The HMS-Sepsis mortality model showed strong discrimination and adequate calibration and reclassified one-third of hospitals to a different performance category from unadjusted mortality. Based on its strong performance, the HMS-Sepsis mortality model can aid in fair hospital benchmarking, assessment of temporal changes, and observational causal inference analysis.

A novel application of waveform matching algorithm for improving monthly runoff forecasting using wavelet–ML models

ABSTRACT The main goal of this study is to enhance the precision and reliability of monthly runoff forecasts within the complex Navrood watershed, situated in northern Iran. The innovative use of a waveform matching algorithm is a defining feature of this study. This approach is vital in optimizing the selection of the mother wavelet, which is a critical component in wavelet analysis. This is a significant divergence from established techniques in hydrological research, indicating a paradigm change in the area. To thoroughly assess model performance, the Technique for Order of Preference by Similarity to the Ideal Solution (TOPSIS) is applied. This all-encompassing evaluation guarantees not only astounding precision but also a near-perfect fit with the ideal solution. The findings highlight the remarkable precision attained by using the hybrid multiresolution analysis (MRA) methodology. The proposed methodology involves the integration of the maximal overlap discrete wavelet transform (MODWT) with a random forest (RF) model, referred to as MRA–RF. The obtained Nash–Sutcliffe efficiency (NSE) score of 0.94 is noteworthy. Furthermore, the model exhibits a low mean absolute error (MAE) of just 0.36 m3/s, a strong p-factor of 73.5%, and a significant d-factor of 37.9% during extensive testing.

Variational Autoencoders for Generative Drug-Gene Interactions in Periodontal Bone Resorption.

Introduction Periodontal bone resorption is a significant dental problem causing tooth loss and impaired oral function. It is influenced by factors such as bacterial plaque, genetic predisposition, smoking, systemic diseases, medications, hormonal changes, and poor oral hygiene. This condition disrupts bone remodeling, favoring resorptive processes. Variational autoencoders (VAEs) can learn the distribution of drug-gene interactions from existing data, identify potential drug targets, and predict therapeutic effects. This study investigates the generation of drug-gene interactions in periodontal bone resorption using VAEs. Methods A bone resorptive drugs dataset was retrieved from Probes and Drugs and analyzed using Cytoscape (https://cytoscape.org/) and CytoHubba (https://apps.cytoscape.org/apps/cytohubba), powerful tools for studying drug-gene interactions in bone resorption. The dataset was then prepared for matrix representation, with normalized input data. It was subsequently divided into training, validation, and testing sets. We then built an encoder-decoder network, defined a loss function, optimized parameters, and fine-tuned hyperparameters. Using VAEs, we generated new drug-gene interactions, assessed model performance, and visualized the latent space with reconstructed drug-gene interactions for further insights. Results The analysis revealed the top hub genes in drug-gene interactions, including Matrix Metalloproteinase (MMP) 14, MMP 9, HIF1A, STAT1, MAPT, CAS9, MMP2, CASP3, MMP1, and MAK1. The VAE's reconstruction accuracy was measured using mean squared error (MSE), with an average squared difference of 0.077. Additionally, the KL divergence value was 2.349, and the average reconstruction log-likelihood was -246. Conclusion The generative variational encoder model for drug-gene interactions in bone resorption demonstrates high accuracy and reliability in representing complex drug-gene relationships within this context.