High Prediction Accuracy Research Articles

High-precision ethanol concentration microsensor with global spectra aided by the multi-layer perceptron

Whispering gallery mode (WGM) resonators can be used for precision measurement thanks to their high sensitivity, small size, and fast response time. Nevertheless, the design of such sensors is usually achieved by selecting a typical single-mode tracking method, which leads to low utilization of a great deal of information in the resonance spectrum and affects the precision. Here, we use the multi-layer perceptron (MLP) deep learning algorithm to train the global spectra and realize the high-precision measurement of ethanol concentration. Firstly, a large number of transmission spectra of different ethanol concentrations are collected and directly used as the original data sets. Secondly, the MLP algorithm is used for training and testing. Finally, the local feature dimension is extracted from the global features of the spectrum for prediction. The results show that the prediction accuracy of the global spectra sensing is 99.81%, which is 13.02% higher than that of extracting 10 local features. In addition, the prediction accuracy of the MLP is compared with four other commonly used machine learning (ML) algorithms, and the results show that the MLP algorithm has the highest prediction accuracy. Therefore, the high-precision ethanol concentration sensor proposed in this paper opens a new way for intelligent optical micro-resonator sensing.

Convolutional Neural Network for Interface Defect Detection in Adhesively Bonded Dissimilar Structures

This study presents an ultrasonic non-destructive method with convolutional neural networks (CNN) used for the detection of interface defects in adhesively bonded dissimilar structures. Adhesive bonding, as the weakest part of such structures, is prone to defects, making their detection challenging due to various factors, including surface curvature, which causes amplitude variations. Conventional non-destructive methods and processing algorithms may be insufficient to enhance detectability, as some influential factors cannot be fully eliminated. Even after aligning signals reflected from the sample surface and interface, in some cases, due to non-parallel interfaces, persistent amplitude variations remain, significantly affecting defect detectability. To address this problem, a proposed method that integrates ultrasonic NDT and CNN, and which is able to recognize complex patterns and non-linear relationships, is developed in this work. Traditional ultrasonic pulse-echo testing was performed on adhesive structures to collect experimental data and generate C-scan images, covering the time gate from the first interface reflection to the time point where the reflections were attenuated. Two classes of datasets, representing defective and defect-free areas, were fed into the neural network. One subset of the dataset was used for model training, while another subset was used for model validation. Additionally, data collected from a different sample during an independent experiment were used to evaluate the generalization and performance of the neural network. The results demonstrated that the integration of a CNN enabled high prediction accuracy and automation of the analysis process, enhancing efficiency and reliability in detecting interface defects.

Prediction of dialysis adequacy using data-driven machine learning algorithms

Background Adequate delivery of hemodialysis (HD), measured by the spKt/V derived from urea reduction, is an important determinant of clinical outcomes in chronic hemodialysis patients. However, the need for pre- and postdialysis blood samples prevented the assessment of spKt/V in every session. Methods This retrospective single-center study was performed on end-stage renal disease (ESKD) patients aged ≥ 18 years who received standard thrice-weekly chronic HD therapy. Eighty-seven variables, including general, intradialytic, and laboratory variables, were collected from the medical records for analysis. Five steps of preprocessing procedure were deployed to select only the most relevant variables. Six binary classification models were developed to predict whether spKt/V was higher than 1.4. Results A total of 1869 HD sessions from 373 ESKD patients were included in this study. The Random Forest model showed the best prediction for dialysis adequacy, with AUROC scores of 0.860 in the validation dataset and 0.873 in the testing dataset. Notably, an accessible model that solely relied on noninvasively collected general and dialysis-related variables maintained high prediction accuracy, with AUROC scores of 0.854 and 0.868 in the validation and testing datasets, respectively. The five most significant predictive variables were vascular access, gender, body mass index, ultrafiltration volume, and dialysis duration. Conclusion The study results suggest that the development of ML models for accurately predicting dialysis adequacy based on general and intradialytic variables is feasible. These models have the potential to be utilized for noninvasive clinical assessments of dialysis adequacy.

A novel operational reliability and state prediction method based on degradation hidden Markov model with random threshold

AbstractBearings are widely used as a common mechanical component and significantly impacts the reliability and safety of various engineering equipment. In the field of large‐scale equipment such as wind turbines, statistical data for reliability and state prediction of bearings is usually limited, leading to the inapplicability of traditional statistical methods. Besides, individual differences commonly exist among bearings, and the appropriate failure threshold for a specific bearing is difficult to be determined due to the individual uncertainty, which may induce prediction errors. To address these issues, a novel operational reliability and state prediction method based on random threshold degradation hidden Markov model was proposed in the study. The traditional hidden Markov model was improved by considering the effect of performance degradation on state transition probability matrix. Moreover, random failure thresholds of performance degradation were employed to describe individual differences between bearings. The operational reliability and state of bearings was obtained by comprehensive analysis of prediction results under different failure thresholds. Verification of the proposed operational reliability and state prediction method were conducted using PHM2012 bearing operation data. The results underscored the effectiveness of the proposed method in achieving a comparatively high accuracy in operational reliability prediction without the prerequisite of an exact failure threshold, when contrasted with several established reliability and life prediction techniques.

Integrating fast iterative filtering and ensemble neural network structure with attention mechanism for carbon price forecasting

Accurate carbon price forecasts are crucial for policymakers and enterprises to understand the dynamics of carbon price fluctuations, enabling them to formulate informed policies and investment strategies. However, due to the non-linear and non-stationary nature of carbon price, traditional models often struggle to achieve high prediction accuracy. To address this challenge, this study proposes a novel integrated prediction framework designed to enhance forecast accuracy. First, the carbon price series is decomposed into a series of smoother subsequences using fast iterative filtering (FIF). Subsequently, an integrated prediction model, AM-TCN-LSTM, is constructed, incorporating the attention mechanism (AM), temporal convolutional networks (TCN), and long short-term memory (LSTM) neural networks. The attention mechanism adaptively captures complex features from multiple factors, while the TCN-LSTM efficiently extracts temporal features from the sequences. Finally, the results from each subsequence are aggregated to generate the final prediction. Five carbon markets in china: Guangdong, Hubei, Shenzhen, Beijing, and Shanghai were selected to verify the validity of the proposed model. Various comparative models and evaluation metrics were employed to assess performance. The results demonstrate that: (1) the TCN-LSTM model achieves higher prediction accuracy compared to single models. (2) FIF is a more effective decomposition method with superior performance compared to EMD-based methods. (3) The proposed model exhibits the highest predictive capability, with MAE values of 0.0964, 0.1403, 1.9476, 2.0848, and 0.5029 for the five carbon markets, significantly outperforming comparison models. (4) The attention mechanism effectively captures the influence of multiple factors on carbon price, particularly within the short-term components.

Underlying factors for gold price prediction based on ARIMA models

Abstract. The purpose of this study is to explore the application of ARIMA model and deep learning technology in time series diagram prediction and the comparison of their effects. First, this paper analyzes the time series data in detail and models them using the ARIMA model. The ARIMA model effectively captures the seasonal and trending characteristics of the data through autoregressive, moving average, and differential steps. At the same time, the method of introducing deep learning models based on recurrent neural networks (RNNs), especially long short-term memory networks (LSTMs), to deal with the nonlinear characteristics and long-term dependency of time series data. By comparing and analyzing the performance of the ARIMA model and the deep learning model in terms of prediction accuracy, computational efficiency and model generalization ability, finding that the deep learning model has higher prediction accuracy when processing complex time series data. In addition, this study also uses time series diagrams to visually display the prediction results, which further verifies the advantages of deep learning models in capturing dynamic changes in data. This study provides a new perspective and method for time series data analysis and prediction, and has important practical significance for finance, meteorology, energy and other fields.

Remote Kinematic Analysis for Mobility Scooter Riders Leveraging Edge AI

Current kinematic analysis for patients with upper or lower extremity challenges is usually performed indoors at the clin- ics, which may not always be accessible for all patients. On the other hand, mobility scooter is a popular assistive tool used by people with mobility disabilities. In this study, we introduce a remote kinematic analysis system for mobility scooter riders to use in their local communities. In order to train the human pose estimation model for the kinematic anal- ysis application, we have collected our own mobility scooter riding video dataset which captures riders’ upper-body move- ments. The ground truth data is labeled by the collaborating clinicians. The evaluation results show high system accuracy both in the keypoints prediction and in the downstream kine- matic analysis, compared with the general-purpose pose mod- els. Our efficiency test results on NVIDIA Jetson Orin Nano also validate the feasibility of running the system in real-time on edge devices.

Prospectively predicting 6-month risk for non-suicidal self-injury among adolescents after psychiatric hospitalization based on a predictive model

BackgroundIt is challenging to predict the occurrence of non-suicidal self-injury (NSSI) among adolescents over short periods. Moreover, the predictive value of indices for NSSI remains elusive. Thus, this study aimed to identify predictors of NSSI within 6 months among adolescents after psychiatric hospitalization by establishing a risk assessment model.MethodsA total of 632 high-risk participants were included in this study. The distribution characteristics of adolescent NSSI were initially assessed through a cross-sectional survey, following which risk factors were identified using logistic regression analysis. The risk score method was then used to construct a 6-month risk assessment model for NSSI. Lastly, the predictive effect of the model was evaluated by indicators such as the area under the receiver operating characteristic (ROC) curve and the positive predictive value.ResultsAfter 6 months, 412 cases of NSSI were identified. According to the logistic regression model, the frequency of relapses, medication status, and NSSI history were identified as influencing factors. Higher scores on the Impulsive Behavior Scale and Pittsburgh Sleep Quality Index were associated with a higher risk of NSSI. Conversely, higher scores on the Pain and Belief Perception Scale were correlated with a lower risk of NSSI. Moreover, the area under the ROC curve for the predictive model was 0.9989, with a 95% confidence interval of (0.9979, 0.9999), highlighting its high predictive ability and accuracy. The predictive model was validated using 78 patients, yielding an area under the ROC curve of 0.9703 and a 95% confidence interval of (0.9167, 0.9999), demonstrating outstanding predictability.ConclusionThese results collectively showed that the predictive model could accurately predict adolescent NSSI. Thus, the model’s primary variables may be applied to predict the risk of NSSI in the clinical setting.

Lactate dehydrogenase A is a diagnostic biomarker associated with immune infiltration, m6A modification and ferroptosis in endometrial cancer

BackgroundLactate dehydrogenase A (LDHA) has been confirmed as a tumor promoter in various cancers, but its role in endometrial cancer remains unclear.MethodsThe Cancer Genome Atlas (TCGA), quantitative real-time polymerase chain reaction and the Human Protein Atlas were utilized to analyzed the LDHA expression in EC. The LDHA levels of patients with different clinical features were compared based on the TCGA cohort. The Genome Ontology, Kyoto Encyclopedia of Genes and Genomes, and Gene Set Enrichment Analysis of LDHA-related genes were conducted by R language. The influence of LDHA knockdown on cell proliferation, apoptosis, migration and invasion was detected by in vitro experiment. The relationship between LDHA expression and immune infiltration was explored by Tumor Immune Estimation Resource 2.0 and Gene Expression Profiling Interactive Analysis. The association of LDHA level with N6-methyladenosine (m6A) modification and ferroptosis was investigated based on the TCGA-UCEC and the GEO cohort.ResultsThe LDHA was overexpressed in EC tissues and EC cell lines, and had high predictive accuracy for the EC diagnosis. The LDHA level was associated with age, histological type, histologic grade, and radiation therapy. LDHA-related genes participated in multiple biological functions and signaling pathways. LDHA downregulation significantly promoted cell apoptosis and inhibited the proliferation, migration, and invasion of EC cells. LDHA expression was connected to multiple tumor-infiltrating lymphocytes (TILs), m6A-related genes, and ferroptosis-related genes.ConclusionLDHA has the potential to work as an EC biomarker associated with TILs, m6A modification, and ferroptosis in EC.

Prognostic nomograms for locally advanced cervical cancer based on the SEER database: Integrating Cox regression and competing risk analysis.

Locally advanced cervical carcinoma (LACC) remains a significant global health challenge owing to its high recurrence rates and poor outcomes, despite current treatments. This study aimed to develop a comprehensive risk stratification model for LACC by integrating Cox regression and competing risk analyses. This was done to improve clinical decision making. We analyzed data from 3428 patients with LACC registered in the Surveillance, Epidemiology, and End Results program and diagnosed them between 2010 and 2015. Cox regression and competing risk analyses were used to identify the prognostic factors. We constructed and validated nomograms for overall survival (OS) and disease-specific survival (DSS). Multivariate Cox regression identified key prognostic factors for OS, including advanced International Federation of Gynecology and Obstetrics stage, age, marital status, ethnicity, and tumor size. Notably, International Federation of Gynecology and Obstetrics stages IIIA, IIIB, and IVA had hazard ratios of 2.227, 2.451, and 4.852, respectively, significantly increasing the mortality risk compared to stage IB2. Ethnic disparities were evident, with African Americans facing a 39.8% higher risk than Caucasians did. Competing risk analyses confirmed the significance of these factors in DSS, particularly tumor size. Our nomogram demonstrated high predictive accuracy, with area under the curve values ranging from 0.706 to 0.784 for DSS and 0.717 to 0.781 for OS. Calibration plots and decision curve analyses further validated the clinical utility of this nomogram. We present effective nomograms for LACC risk stratification that incorporate multiple prognostic factors. These models provide a refined approach for individualized patient management and have the potential to significantly enhance therapeutic strategies for LACC.

Machine learning insights into regional dynamics and prevalence of COVID-19 variants in US health and human services regions

BackgroundThe COVID-19 pandemic arising from the emergence of SARS-CoV-2 in late 2019 has led to global devastation with millions of lives lost by January 2024. Despite the WHO’s declaration of the end of the global health emergency in May 2023, the virus persists, propelled by mutations. Variants continue to challenge vaccination efforts, underscoring the necessity for ongoing vigilance. This study aimed at contributing to a more data-driven approach to pandemic management by employing random forest regression to analyze regional variant prevalence.MethodsThis study utilized data from various sources including National COVID Cohort Collaborative database, Bureau of Transportation Statistics, World Weather Online, EPA, and US Census. Key variables include pollution, weather, travel patterns, and demographics. Preprocessing steps involved merging and normalization of datasets. Training data spanned from January 2021 to February 2023. The Random Forest (RF) Regressor was chosen for its accuracy in modeling. To prevent data leakage, time series splits were employed. Model performance was evaluated using metrics such as MSE and R-squared.ResultsThe Alpha variant was predominant in the Southeast, with less than 80% share even at its peak. Delta surged initially in Kansas City and maintained dominance there for over 5 months. Omicron subvariant BA.5 spread nationwide, becoming predominant across all Health and Human Services regions simultaneously, with New York seeing the earliest and fastest decline in its share. Variant XBB.1.5 concentrated more in the Northeast, but limited data hindered full analysis. Using RF Regressor, key features affecting spread patterns were identified, with high predictive accuracy. Each variant showed specific environmental correlations; for instance, Alpha with air quality index, Delta with ozone density, BA.5 with UV index, and XBB.1.5 with land area and income. Correlation analysis further highlighted variant-specific associations.ConclusionsThis research provides a comprehensive analysis of the regional distribution of COVID-19 variants, offering critical insights for devising targeted public health strategies. By utilizing machine learning, the study uncovers the complex factors contributing to variant spread and reveals how specific factors contribute to variant prevalence, offering insights crucial for pandemic management.

Indoor Air Quality Monitoring System with High Accuracy of Gas Classification and Concentration Prediction via Selective Mechanism Research.

The efficacy of sensors, particularly sensor arrays, lies in their selectivity. However, research on selectivity remains notably obscure and scarce. In this work, indoor pollutants (C7H8, HCHO, CH4, and NO2) were chosen as the target gas. Following the screening of six oxides from previous work, temperature-programmed desorption/reduction experiments were conducted to delve into the origins of selectivity. The results explicate the superiority of NiO in detecting toluene and unveil the distinctive NO2 sensing mechanism of WO3 sensors. Based on the sensor array comprising these oxides, it can clearly detect low concentrations of C7H8 (S = 1.6 to 50 ppb), HCHO (S = 1.4 to 50 ppb), and NO2 (S = 3.3 to 50 ppb), which satisfies the requisites of indoor air monitoring. Meanwhile, three machine learning models (Extreme Gradient Boosting, Support Vector Machine, and Back Propagation Neural Network) are employed for gas classification. The classification accuracies of these models are 95.45%, 100%, and 100%, while the R2 values of the concentration prediction are 99.65%, 94.9%, and 98.04%, respectively, indicating the rationality of material selection. Furthermore, it can still achieve relatively high accuracy in gas classification (94.12%) and concentration prediction (89.36%), even for gas mixtures of four gases. Finally, an indoor air quality monitoring system is developed, which enables real-time monitoring of indoor gas quality through the Internet of Things.

Cross-instrument optical coherence tomography-angiography (OCTA)-based prediction of age-related macular degeneration (AMD) disease activity using artificial intelligence

This study investigates the efficacy of predicting age-related macular degeneration (AMD) activity through deep neural networks (DNN) using a cross-instrument training dataset composed of Optical coherence tomography-angiography (OCTA) images from two different manufacturers. A retrospective cross-sectional study analyzed 2D vascular en-face OCTA images from Heidelberg Spectralis (1478 samples: 1102 training, 276 validation, 100 testing) and Optovue Solix (1003 samples: 754 training, 189 validation, 60 testing). OCTA scans were labeled based on clinical diagnoses and adjacent B-scan OCT fluid information, categorizing activity into normal, dry AMD, active wet AMD, and wet AMD in remission. Experiments explored cross-instrument disease classification using separate and combined datasets for training the DNN. Testing involved 100 Heidelberg and 60 Optovue samples. Training on Heidelberg data alone yielded 73% accuracy on Heidelberg images and 60% on Optovue images. Training on Optovue data alone resulted in 34% accuracy on Heidelberg and 85% on Optovue images. Combined training data from both instruments achieved 78% accuracy on Heidelberg and 76% on Optovue test sets. Results indicate that cross-instrument classifier training demonstrates high classification prediction accuracy, making cross-instrument training viable for future clinical applications. This implies that vascular morphology in OCTA can predict disease progression.

Prediction of Healing Trajectory of Chronic Wounds Using a Machine Learning Approach.

Objective: New treatment options are emerging for chronic wounds, which represent a growing problem because of population ageing and increasing burden of chronic disease. While promising, the existing evidence for advanced modalities is commonly derived from small and/or poorly controlled studies and clear criteria for selecting patients, who are likely to benefit from these expensive options are lacking. In this study, we develop and validate a machine learning model to predict if a chronic wound, independent of etiology, is expected to heal within 12 weeks to identify cases in potential need of advanced treatment options. Approach: Retrospective analysis of electronic health record data from 2014 to 2018 covering 532 wound care clinics in the United States and 261,398 patients with 620,356 unique wounds. Prediction of 12-week healing trajectories with a machine learning model. Results: The best-performing model in a training dataset of a randomly drawn 75% subset of wounds contained variables for patient demographics, comorbidities, wound characteristics at initial presentation, and changes in wound dimensions over time, with the latter group being the most influential predictors. The final machine learning model had a high predictive accuracy with area under the receiver operating characteristic curves of 0.9 and 0.92 after 4 and 5 weeks of treatment, respectively. Innovation: A machine learning model can identify chronic wounds at risk of not healing by week 12 with high accuracy in the early weeks of treatment. Conclusions: If embedded in real-world care, the generated information could be able to guide effective and efficient treatment decisions.

Study on Strength Prediction and Material Scheme Optimization for Modified Red Mud Based on Artificial Neural Networks

Focusing on the complex nonlinear problems of strength prediction and the material scheme design of modified red mud for use as a road material in engineering applications, a strength prediction neural network is established and utilized to optimize the material scheme, including the compound-solidifying agent ratio, water content, and curing age, based on experimental data accumulated during years of engineering practice and an artificial neural network. In this study, a backpropagation (BP) neural network is adopted, and 114 sets of experimental data are used to train the parameters of the unconfined compressive strength prediction model. Then, using the BP strength prediction model, the material scheme optimization process is carried out, with the strength and material costs as the objectives. The results show that the BP neural network model has a high prediction accuracy, the relative prediction error is basically within 10%, the root-mean-squared error is less than 0.04, and the correlation coefficient is more than 0.99. According to the strength requirements of modified red mud in different road projects and the constraints of each property, an optimal material scheme with a lower cost and higher 7 d target strength is obtained using a mix of polymer agent–fly-ash–cement–speed-cement in a ratio of 0.02%:1.96%:4.78%:0%, with a 33.93% water content of raw red mud, so that the target strength and material cost are 2.987 MPa and 17.099 CNY/T. This study creates an optimal material scheme, incorporating the compound-solidifying agent ratio, curing age, and water content of the modified red mud road material according to the strength requirements of different projects, thereby promoting the popularization of the utilization of red mud with better engineering practicability and economy.

Concurrent oculomotor hyperactivity and deficient anti-saccade performance in obsessive-compulsive disorder

Existing studies mainly focused on the inhibition of the task-interfering response to understand the inhibitory deficits of obsessive-compulsive disorder (OCD). However, recent studies suggested that inhibitory function is broadly involved in response preparation and implementation. It is yet unknown if the inhibition dysfunction in OCD extends beyond the task-interfering response to the general inhibitory function. Here we address this issue based on the multidimensional eye-movement measurements, which can better capture the inhibitory deficits than manual responses. Thirty-one OCD patients and 32 healthy controls (HCs) completed the anti-saccade task where multidimensional eye-movement features were developed. Confirmatory factor analysis (CFA) suggested two components of inhibitory function that negatively correlated with each other: one component of oculomotor hyperactivity in generating oculomotor output which is characterized with early premature saccades, early cross rates and saccade number; the other component of task-specific oculomotor efficiency which is characterized with task accuracy, saccade latency, correction rate, and amplitude gain. Importantly, OCD showed both stronger oculomotor hyperactivity and deficient oculomotor efficiency than HCs, and the machine-learning-based classifications showed that the features of oculomotor hyperactivity had higher prediction accuracy than the features of oculomotor efficiency in distinguishing OCD from HCs. Our results suggested that OCD has concurrent deficits in oculomotor hyperactivity and oculomotor efficiency, which may originate from a common inhibitory dysfunction.

Performance of uncertainty-based active learning for efficient approximation of black-box functions in materials science.

Obtaining a fine approximation of a black-box function is important for understanding and evaluating innovative materials. Active learning aims to improve the approximation of black-box functions with fewer training data. In this study, we investigate whether active learning based on uncertainty sampling enables the efficient approximation of black-box functions in regression tasks using various material databases. In cases where the inputs are provided uniformly and defined in a relatively low-dimensional space, the liquidus surfaces of the ternary systems are the focus. The results show that uncertainty-based active learning can produce a better black-box function with higher prediction accuracy than that by random sampling. Furthermore, in cases in which the inputs are distributed discretely and unbalanced in a high-dimensional feature space, datasets extracted from materials databases for inorganic materials, small molecules, and polymers are addressed, and uncertainty-based active learning is occasionally inefficient. Based on the dependency on the material descriptors, active learning tends to produce a better black-box functions than random sampling when the dimensions of the descriptor are small. The results indicate that active learning is occasionally inefficient in obtaining a better black-box function in materials science.

Boosting Point Set-Based Network with Optimal Transport Optimization for Oriented Object Detection

When handling complex remote sensing scenarios, rotational angle information can improve detection accuracy and enhance algorithm robustness, providing support for fine-grained detection. Point set representation is one of the most commonly used methods in arbitrary-oriented object detection tasks, leveraging discrete feature points to represent oriented targets and achieve high accuracy in angle prediction. However, due to the inherent discreteness of point set representation, it is prone to significant impact from isolated points and representational ambiguity in harsh application scenarios, leading to inaccurate detection. To address this issue, an efficient aerial object detector named BE-Det is proposed, which uses the optimal transport (OT) strategy to constrain the positions of isolated points. Additionally, a candidate point set quality evaluation scheme is designed to effectively assess the quality of candidate point sets. Experimental results on two challenging aerial datasets demonstrate that the proposed method outperforms several advanced detection methods.

AI-Enhanced Prediction of Pavement Crack Propagation: A Study Using Traffic Load, Environmental and Material Data

This study develops an AI-based predictive model for forecasting pavement crack propagation by integrating traffic load, environmental conditions, and material property data. Traditional pavement management systems often struggle to accurately predict crack growth due to the complex interactions between these influencing factors. By leveraging data from various sources, including sensor-based traffic metrics, meteorological data, and material composition tests, this study identifies significant variables contributing to crack initiation and progression. The proposed model utilizes a blend of machine learning algorithms, including Random Forest and neural networks, with a cross-validation approach to ensure robustness. Results indicate that the model achieves high prediction accuracy, with an RMSE of 1.2 mm/year and an R-squared value close to 0.93. The findings support the use of AI-enhanced models as reliable tools for road infrastructure planning and maintenance, promising reductions in maintenance costs and improved pavement durability.

Random survival forest algorithm for risk stratification and survival prediction in gastric neuroendocrine neoplasms.

This study aimed to construct and assess a machine-learning algorithm designed to forecast survival rates and risk stratification for patients with gastric neuroendocrine neoplasms (gNENs) after diagnosis. Data on patients with gNENs were extracted and randomly divided into training and validation sets using the Surveillance, Epidemiology, and End Results database. We developed a prediction model using 10 machine learning algorithms across 101 combinations to forecast cancer-related mortality in patients with gNENs, selecting the best model using the highest mean over a sequence of time-dependent area under the receiver operating characteristic (ROC) curve (AUC). The performance of the final model was assessed through time-dependent ROC curves for discrimination and calibration curves for calibration. The maximum selection rank method was used to determine the best prognostic risk score threshold for classifying patients into high- and low-risk groups. Afterward, Kaplan-Meier analysis and log-rank test were used to compare survival rates among these groups. Our study examined 775 patients with gNENs, dividing them into training and validation sets. A training set comprised 543 patients, with a median follow-up of 42 months and cumulative mortality rates of 40.0% at 1 year, 48.6% at 3 years, and 54.0% at 5 years. A validation set comprised 232 patients, with cumulative mortality rates of 29.1% at 1 year, 43.5% at 3 years, and 53.2% at 5 years. The optimal random survival forest (RSF) model (mtry = 4, node size = 5) achieved an AUC of 0.839 for survival prediction in the training set. Comprising 11 variables such as demographics, treatment details, tumor characteristics, T staging, N staging, and M staging, the RSF model revealed high predictive accuracy with AUCs of 0.92, 0.96, and 0.96 for 1-, 3-, and 5-year survival, respectively, which was consistently reflected in the validation set with AUCs of 0.88, 0.92, and 0.89, respectively. Moreover, patients were risk-stratified. Although our RSF model effectively stratified patients into different prognostic groups, it needs external validation to confirm its utility for noninvasive prognostic prediction and risk stratification in gNENs. Further research is required to verify its broader clinical applicability.