Traditional Statistical Models Research Articles

Sample size and predictive performance of machine learning methods with survival data: A simulation study.

Prediction models are increasingly developed and used in diagnostic and prognostic studies, where the use of machine learning (ML) methods is becoming more and more popular over traditional regression techniques. For survival outcomes the Cox proportional hazards model is generally used and it has been proven to achieve good prediction performances with few strong covariates. The possibility to improve the model performance by including nonlinearities, covariate interactions and time-varying effects while controlling for overfitting must be carefully considered during the model building phase. On the other hand, ML techniques are able to learn complexities from data at the cost of hyper-parameter tuning and interpretability. One aspect of special interest is the sample size needed for developing a survival prediction model. While there is guidance when using traditional statistical models, the same does not apply when using ML techniques. This work develops a time-to-event simulation framework to evaluate performances of Cox regression compared, among others, to tuned random survival forest, gradient boosting, and neural networks at varying sample sizes. Simulations were based on replications of subjects from publicly available databases, where event times were simulated according to a Cox model with nonlinearities on continuous variables and time-varying effects and on the SEER registry data.

Forecasting the Acute Heart Failure Admissions: Development of Deep Learning Prediction Model Incorporating the Climate Information

BackgroundClimate is known to influence the incidence of cardiovascular events. However, their prediction with traditional statistical models remain imprecise. Methods and ResultsWe analyzed 27,799 acute heart failure (AHF) admissions within the Tokyo CCU Network Database from January 2014 to December 2019. High-risk AHF (HR-AHF) day was defined as a day with the upper 10th percentile of AHF admission volume. Deep neural network (DNN) and traditional regression models were developed using the admissions in 2014-2018 and tested in 2019. Explanatory variables included 17 meteorological parameters. Shapley additive explanations were used to evaluate their importance. The median number of incidences of AHF was 12 (9-16) per day in 2014–2018 and 11 (9-15) per day in 2019. The predicted AHF admissions correlated well with the observed numbers (DNN: R2 = 0.413, linear regression: R2 = 0.387). The DNN model was superior in predicting HR-AHF days compared with the logistic regression model [c-statistics: 0.888 (95%CI: 0.818-0.958) vs. 0.827 (95%CI: 0.745-0.910): p=0.0013]. Notably, the strongest predictive variable was the 7-day moving average of the lowest ambient temperatures. ConclusionsThe DNN model had good prediction ability for incident AHF using climate information. Forecasting AHF admissions could be useful for the effective management of AHF.

Predicting success of atrial fibrillation ablation: comparing machine learning approaches of intracardiac electrograms

Abstract Introduction Machine learning (ML) and Deep learning (DL) neural networks have been used to predict which patients will respond to atrial fibrillation (AF) ablation. However, the predictive results of DL are modest compared to traditional statistical models of clinical features with or without intracardiac electrograms. Purpose To test the hypothesis that optimized ML architectures will better identify patients who respond to AF ablation using intracardiac EGM than other DL architectures in our large registry of 320 consecutive patients. Methods Our patients had age 65.1±10.4Y, were 25% women, 61.6% non-paroxysmal, with intracardiac EGM recorded from multipolar catheters and 64-pole baskets (Fig A). Patients were propensity matched into those in whom ablation terminated by ablation (N=160, "Term") and those without termination (N=160, "Non-Term"). Using a proven DL architecture, ResNet, we systematically varied the depth (i.e., number of residual layers (D)) and width (i.e., number of nodes per layer (W)). We predicted termination (Fig C), presenting EGMs to training cohorts and testing in separate cohorts in a 6-fold cross-validation which we repeated 10 times for each fold with different randomization. Results Fig B illustrates that the prediction accuracy is generally negatively correlated with the network depth but positively correlated with the width. Specifically, the deep architecture (D=4, W=256, with 265M trainable parameters, blue-circled in Fig B) predicted with AUC = 0.61±0.08. However, a shallow-narrow architecture (D=1, W=16, green-circled in Fig B) achieved same AUC (0.61±0.09, p=0.65), while reducing the number of parameters by 97% compared to the DL architecture (79K vs. 265M). Additionally, a shallow-wide architecture (D=1, W=2048, red-circled in Fig B) achieved higher AUC (0.74±0.08; p<0.0001), while saving 44% of the parameters (148M vs 265M) and 26% training time compared to the deep architecture. Conclusion Shallow neural network classification of raw EGMs predicted ablation response with better accuracy than other architectures. Future work should determine if deep architectures ultimately ‘overfit’ detailed EGM data, which may explain their lower success.Figure

Intra-day solar irradiation forecast using machine learning with satellite data

Northwest China is rich in solar resource and the photovoltaic manufacturing industry is developing rapidly. Accurate solar radiation forecast suitable for the northwest desert area has become an urgent need. In this work, we use the classical machine learning models (SVM and RF) that are more complex than traditional statistical models to forecast intra-day global solar irradiance (GHI). Since there are some uncertainties in solar radiation attenuation models. We explore an approach that requires little preprocessing to enter satellite data as input: the mean of the satellite image window. Such a process provides a direct GHI forecast without using the clear sky index as a proxy. The model includes several satellite channels, not only visible channels. Since China's Fengyun-4 series satellites (FY-4) are the new generation of stationary meteorological satellite and have not yet been fully tested and applied to solar irradiance prediction, the regional average of each channels (Channel01∼Channel07) of FY-4A satellite cloud image are taken as important parameters input of the ML model, with lead time from 10 min to 3 h. The combination of climatology and persistence (Clim-Pers) model is chosen as the benchmark model. Our cases studies in Yuzhong, Minqin and Dunhuang show that the FS of RF model is higher than the SVM model in all forecast cases, and the performance advantage becomes more obvious when the lead time beyond 90 min. The FS values of RF model in Yuzhong, Minqin and Dunhuang site at time horizons 10 min-3 h are 13.5–37.6%, 18.2–35.8% and 17.3–34.2%, respectively, the forecast performance is very stable in different climate types. Therefore, it is a good and simple way to improve the accuracy of ultra-short-term solar forecasting by introducing satellite observations into the ML model.

Retrieval of suspended sediment concentrations using remote sensing and machine learning methods: A case study of the lower Yellow River

Monitoring suspended sediment concentrations (SSCs) of channel waters is essential for evaluating important issues such as erosion, sediment transport and deposition, reservoir siltation, and water pollution throughout watersheds. Remote sensing technology provides a new approach to large-scale SSC monitoring, but the accuracy of existing retrieval methods based on remote sensing technology still needs to be improved for complex environments. This study proposes a new hyperparameter and globally adaptive retrieval (HGAR) method for river SSC retrieval that combines a remote sensing technique and machine learning methods. The HGAR method uses the light gradient-boosting machine method as the basic model for SSC retrieval, and the tree-structured Parzen estimator method, coefficient of determination, and residual prediction deviation are combined to optimize the basic model’s hyperparameters. The experiment was conducted in the lower Yellow River, where 24,918 Landsat images from 1997 to 2020 were acquired using Google Earth Engine, and 1,772 monthly cloud-free images were obtained in the end. The sequential backward floating selection method was adopted to determine the optimal modeling features. Two regression and three machine learning methods were compared with the proposed method. The results demonstrate that the proposed HGAR method significantly outperforms other retrieval methods in terms of retrieval accuracy (as evidenced by a correlation coefficient of 0.626 and a Nash-Sutcliffe efficiency coefficient of 0.613), generalizability, and robustness. Finally, the model explanation results show that the HGAR method can successfully capture the key information for SSC retrieval and has higher flexibility than traditional statistical models. In addition, we found that the overall fit and the fit of local extreme values are inconsistent in the lower Yellow River, and the proposed improved accuracy indicator balances the two aspects well. In conclusion, the proposed HGAR method can be used effectively to estimate channel SSC and is expected to provide technical support for long-term and large-scale monitoring of the channel surface SSC.

Application of Machine Learning Based on Structured Medical Data in Gastroenterology.

The era of big data has led to the necessity of artificial intelligence models to effectively handle the vast amount of clinical data available. These data have become indispensable resources for machine learning. Among the artificial intelligence models, deep learning has gained prominence and is widely used for analyzing unstructured data. Despite the recent advancement in deep learning, traditional machine learning models still hold significant potential for enhancing healthcare efficiency, especially for structured data. In the field of medicine, machine learning models have been applied to predict diagnoses and prognoses for various diseases. However, the adoption of machine learning models in gastroenterology has been relatively limited compared to traditional statistical models or deep learning approaches. This narrative review provides an overview of the current status of machine learning adoption in gastroenterology and discusses future directions. Additionally, it briefly summarizes recent advances in large language models.

Prediction for the Sluice Deformation Based on SOA-LSTM-Weighted Markov Model

Increasingly, deformation prediction has become an essential research topic in sluice safety control, which requires significant attention. However, there is still a lack of practical and efficient prediction modeling for sluice deformation. In order to address the limitations in mining the deep features of long-time data series of the traditional statistical model, in this paper, an improved long short-term memory (LSTM) model and weighted Markov model are introduced to predict sluice deformation. In the method, the seagull optimization algorithm (SOA) is utilized to optimize the hyper-parameters of the neural network structure in LSTM primarily to improve the model. Subsequently, the relevant error sequences of the fitting results of SOA-LSTM model are classified and the Markovity of the state sequence is examined. Then, the autocorrelation coefficients and weights of each order are calculated and the weighted and maximum probability values are applied to predict the future random state of the sluice deformation. Afterwards, the prediction model of sluice deformation on the SOA-LSTM-weighted Markov model is proposed. Ultimately, the presented model is used to predict the settlement characteristics of an actual sluice project in China. The analysis results demonstrate that the proposed model possesses the highest values of R2 and the smallest values of RMSE and absolute relative errors for the monitoring data of four monitoring points. Consequently, it concluded that the proposed method shows better prediction ability and accuracy than the SOA-LSTM model and the stepwise regression model.

Monthly Runoff Prediction by Combined Models Based on Secondary Decomposition at the Wulong Hydrological Station in the Yangtze River Basin

Accurate runoff prediction plays a pivotal role in facilitating effective watershed management and the rational allocation of water resources. Given the inherent challenges posed by the nonlinear and nonstationary nature of runoff sequences, this study introduces a novel coupled model, combining complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), variational mode decomposition (VMD), long short-term memory (LSTM), and Informer techniques for monthly runoff prediction at the Wulong hydrological station of the Yangtze River Basin from 1973 to 2022. In addition, by comparing the prediction results of the traditional statistical model, which is the seasonal exponential smoothing model, with those of the machine learning model, the prediction accuracy of the machine learning model was found to be much higher than that of the traditional statistical mode. And the coupled model of secondary decomposition and secondary prediction was compared with other types of coupled models, such as one decomposition and one prediction. The CEEMDAN-VMD-LSTM-Informer model exhibited superior performance, as evidenced by an NSE value of 0.997, MAE of 1.327 × 108 m3, MAPE of 2.57%, and RMSE of 2.266 × 108 m3. The combined model proposed in this paper has the highest prediction accuracy, rendering it suitable for long-time series prediction. Accurate runoff prediction plays a pivotal role in facilitating effective watershed management and the rational allocation of water resources.

Deep learning methods used in movie recommendation systems

As the amount of internet movie data grows rapidly, traditional movie recommendation systems face increasing challenges. They typically rely on statistical algorithms such as item-based or user-based collaborative filtering. However, these algorithms struggle to handle large-scale data and often fail to capture the complexity and contextual information of user behavior. Therefore, deep learning techniques have been widely applied to movie recommendation systems. This paper reviews movie recommendation algorithms based on traditional statistical models and introduces three main deep learning techniques: Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), and Recurrent Neural Networks (RNN). ANN can extract features at different levels of users and movies; CNN can capture features of movie posters and movie data to recommend similar movies; RNN can consider user historical behavior and contextual information to better understand user interests and demands. The application of these deep learning techniques can enhance the accuracy and user experience of movie recommendation systems. This paper also demonstrates the advantages and disadvantages of these models and their specific application methods in movie recommendation systems, and points out the direction for further development and improvement of deep learning models in this field.

Accident severity prediction modeling for road safety using random forest algorithm: an analysis of Indian highways.

Background: Road accidents claim around 1.35 million lives annually, with countries like India facing a significant impact. In 2019, India reported 449,002 road accidents, causing 151,113 deaths and 451,361 injuries. Accident severity modeling helps understand contributing factors and develop preventive strategies. AI models, such as random forest, offer adaptability and higher predictive accuracy compared to traditional statistical models. This study aims to develop a predictive model for traffic accident severity on Indian highways using the random forest algorithm. Methods: A multi-step methodology was employed, involving data collection and preparation, feature selection, training a random forest model, tuning parameters, and evaluating the model using accuracy and F1 score. Data sources included MoRTH and NHAI. Results: The classification model had hyperparameters 'max depth': 10, 'max features': 'sqrt', and 'n estimators': 100. The model achieved an overall accuracy of 67% and a weighted average F1-score of 0.64 on the training set, with a macro average F1-score of 0.53. Using grid search, a random forest Classifier was fitted with optimal parameters, resulting in 41.47% accuracy on test data. Conclusions: The random forest classifier model predicted traffic accident severity with 67% accuracy on the training set and 41.47% on the test set, suggesting possible bias or imbalance in the dataset. No clear patterns were found between the day of the week and accident occurrence or severity. Performance can be improved by addressing dataset imbalance and refining model hyperparameters. The model often underestimated accident severity, highlighting the influence of external factors. Adopting a sophisticated data recording system in line with MoRTH and IRC guidelines and integrating machine learning techniques can enhance road safety modeling, decision-making, and accident prevention efforts.

How does the built environment affect hotel prices? A study using multiscale GWR and deep learning

ABSTRACT This paper utilizes deep learning and street view images to extract visual variables and explore the interactive effects and spatial impacts of the built environment on hotel prices with GeoDetector and multiscale geographically weighted regression (MGWR) models. The results indicate that hotel prices are influenced by multiple factors in a coupled manner, with combined effects more substantial than individual drivers; the visual quality of the built environment can influence consumers’ perception and willingness to stay, thereby affecting hotel prices; MGWR, compared to traditional statistical models, demonstrates superior explanatory power for hotels, accurately identifying the spatial heterogeneity of factors influencing hotel prices; and by elucidating the bandwidth influence and spatial differentiation pattern of each factor, this study provides a reference for hotel site selection, price adjustment, and urban planning.

Quality by Design (QbD) and Design of Experiments (DOE) as a Strategy for Tuning Lipid Nanoparticle Formulations for RNA Delivery.

The successful development of nonviral delivery systems for nucleic acids has been reported extensively over the past years. Increasingly employed to improve the delivery efficiency and therapeutic efficacy of RNA are lipid nanoparticles (LNPs). Many of the various critical formulation parameters can affect the quality attributes and effectiveness of these nano-formulations. Therefore, the systematic drug development approach (QbD) and multivariate design and statistical analysis (DOE) can be very helpful and recommended for the optimization of the composition and production of RNA-LNPs. This review addresses the concepts and applications of QbD and/or DOE for the development of lipid nanoparticles for the delivery of different types of RNA, reporting examples published in the ten recent years presenting the latest trends and regulatory requirements as well as the modern mathematical and statistical design methods. As the topic explored in this review is a novel approach, the full QbD has been described in only a few papers, and a few refer only to some aspects of QbD. In contrast, the DOE approach has been used in most of the optimization works. Different approaches and innovations in DOE have been observed. Traditional statistical tests and modeling (ANOVA, regression analysis) are slowly being replaced by artificial intelligence and machine learning methods.

A novel deep-learning technique for forecasting oil price volatility using historical prices of five precious metals in context of green financing – A comparison of deep learning, machine learning, and statistical models

This study proposes a novel deep-learning convolution neural network (CNN) to forecast crude oil prices based on historical prices of five precious metals (Gold, Silver, Platinum, Palladium, and Rhodium) in context of green financing. The proposed deep learning CNN has three components: a convolution block called a group block, a novel convolutional neural network architecture called GroupNet, and a regression layer. The proposed model is tested against seven machine learning models and three traditional statistical models for predicting oil price volatility using the same independent variables (5 precious metals). A comparison of the deep learning model (our proposed model) with machine learning/deep learning models and statistical methods indicates that the proposed deep learning model has the highest prediction accuracy. A feature selection technique is also applied using the WEKA ML tool to improve the accuracy of the proposed model and existing machine learning and traditional statistical models. The findings indicate a non-linear correlation between oil price volatility and prices of precious metals. Moreover, statistical analysis indicates that deep learning can be used to predict oil price volatility with greater accuracy than machine learning and statistical methods while using precious metals as predictors. The results also indicate that machine learning models (Decision Tables and M5rules) can be used to predict oil price volatility with considerable accuracy. Moreover, the study proves that traditional statistical models can perform better than a few machine learning models (Lazy LWL and GPR).

Deep learning models for price forecasting of financial time series: A review of recent advancements: 2020–2022

AbstractAccurately predicting the prices of financial time series is essential and challenging for the financial sector. Owing to recent advancements in deep learning techniques, deep learning models are gradually replacing traditional statistical and machine learning models as the first choice for price forecasting tasks. This shift in model selection has led to a notable rise in research related to applying deep learning models to price forecasting, resulting in a rapid accumulation of new knowledge. Therefore, we conducted a literature review of relevant studies over the past 3 years with a view to aiding researchers and practitioners in the field. This review delves deeply into deep learning‐based forecasting models, presenting information on model architectures, practical applications, and their respective advantages and disadvantages. In particular, detailed information is provided on advanced models for price forecasting, such as Transformers, generative adversarial networks (GANs), graph neural networks (GNNs), and deep quantum neural networks (DQNNs). The present contribution also includes potential directions for future research, such as examining the effectiveness of deep learning models with complex structures for price forecasting, extending from point prediction to interval prediction using deep learning models, scrutinizing the reliability and validity of decomposition ensembles, and exploring the influence of data volume on model performance.This article is categorized under: Technologies > Prediction Technologies > Artificial Intelligence

A methodological study of exposome based on an open database: Association analysis between exposure to metal mixtures and hyperuricemia

BackgroundExposome recognizes that humans are constantly exposed to multiple environmental factors, and elucidating the health effects of complex exposure mixtures places greater demands on analytical methods. ObjectsWe aimed to explore the association between mixed exposure to metals and hyperuricemia (HUA), and highlight the potential of explainable machine learning (EML) and causal mediation analysis (CMA) for application in the analysis of exposome data. MethodsPre-pandemic data from the National Health and Nutrition Examination Survey (NHANES) 2011–2020 and a total of 13780 individuals were included. We first used traditional statistical models (multiple logistic regression (MLR) and restricted cubic spline regression (RCS)) and EML to explore associations between mixed metals exposures and HUA, followed by the CMA using the 4-way decomposition method to analyze the interaction and mediation effects among BMI or estimated glomerular filtration rate (eGFR), metals and HUA. ResultsThe prevalence of HUA was 18.91% (2606/13780). The MLR showed that mercury (Q4 vs Q1: OR = 1.08, 95% CI:1.02–1.14) and lead (Q4 vs Q1: OR = 1.23, 95% CI:1.13–1.34) were generally positively associated with HUA. Higher concentrations of lead, mercury, selenium and manganese were associated with the increased odds of HUA, and BMI and eGFR were the top two variables attributable to the risk of developing HUA in the EML. Subgroup analyses from the MLR and EML consistently demonstrated the positive relationship between exposure to lead, mercury and selenium in participants with BMI <25 kg/m2 and BMI ≥30 kg/m2. BMI mediated 32.12% of the association between lead exposure and HUA, and the interaction between BMI and lead accounted for 3.88% of the association in the CMA. ConclusionsHeavy metals can increase the HUA risk and BMI or eGFR can mediate and interact with metals to cause HUA. Future studies based on exposome can attempt to utilize the EML and CMA.

Artificial neural networks for predicting mechanical properties of Al2219-B4C-Gr composites with multireinforcements

Artificial neural networks (ANNs) have gained prominence as a reliable model for clustering, grouping, and analysis in various domains. In recent times, machine learning (ML) models such as ANNs have proved to be on par with traditional regression and statistical models in terms of performance and usability. This study focuses on the fabrication of multicomponents-reinforced composites (Boron carbide (B4C) and Graphite (Gr)) using the stir casting technique. The addition of Magnesium to the melt enhances the wettability of B4C and Gr particles within the matrix. The microstructure and mechanical properties of the resulting Al-Mg-metal matrix composites (MMCs) are analyzed. Scanning electron micrographs reveal that B4C and Gr particles were uniformly dispersed in the matrix. X-Ray diffraction analysis confirmed the dispersion of the strengthening. The mechanical properties, including hardness, tensile, compressive, and impact strength, increased with the increase in B4C and Gr wt.%. As the percentage of B4C and Gr reinforcement wt.% increased, the load on the matrix reduced and its load-bearing capacity improved. The strain field generation rate also increased with an increase in B4C and Gr in the matrix, resulting in enhanced mechanical properties. The ANN analysis further confirmed that B4C was the more significant contributor to the mechanical properties of the composites.

Deep Learning-Enabled Statistical Model Estimation for Power Transformers with Censoring and Truncation Problems

Traditional statistical models, e.g., Weibull distributions, are popular solutions for failure modeling and degradation anal- ysis in a variety of industries. To estimate the parameters of these statistical models, maximum likelihood estimation (MLE) is often engaged through various optimization algo- rithms. However, when dealing with highly reliable or new equipment, it is challenging to fit limited or unbalanced data to obtain an accurate model. In this paper, we propose a deep learning (DL)-based model for estimating the Weibull param- eters with both censoring and truncation problems. Instead of using the conventional matrices such as concordance index, we propose a novel validation framework to examine the pre- diction accuracy of different models. We examine the perfor- mance of the proposed approach on real-world power trans- former data, and the results show that our approach can im- prove prediction accuracy and is less susceptible to the trun- cation problem. Our results also suggest that deep learning techniques can help enhance traditional statistical modeling for reliability analysis.

Machine learning applications and challenges in graft-versus-host disease: a scoping review.

This review delves into the potential of artificial intelligence (AI), particularly machine learning (ML), in enhancing graft-versus-host disease (GVHD) risk assessment, diagnosis, and personalized treatment. Recent studies have demonstrated the superiority of ML algorithms over traditional multivariate statistical models in donor selection for allogeneic hematopoietic stem cell transplantation. ML has recently enabled dynamic risk assessment by modeling time-series data, an upgrade from the static, "snapshot" assessment of patients that conventional statistical models and older ML algorithms offer. Regarding diagnosis, a deep learning model, a subset of ML, can accurately identify skin segments affected with chronic GVHD with satisfactory results. ML methods such as Q-learning and deep reinforcement learning have been utilized to develop adaptive treatment strategies (ATS) for the personalized prevention and treatment of acute and chronic GVHD. To capitalize on these promising advancements, there is a need for large-scale, multicenter collaborations to develop generalizable ML models. Furthermore, addressing pertinent issues such as the implementation of stringent ethical guidelines is crucial before the widespread introduction of AI into GVHD care.

Prediction Model for Postoperative Quality of Life Among Breast Cancer Survivors Along the Survivorship Trajectory From Pretreatment to 5 Years: Machine Learning-Based Analysis.

Breast cancer is the most common cancer and the most common cause of cancer death in women. Although survival rates have improved, unmet psychosocial needs remain challenging because the quality of life (QoL) and QoL-related factors change over time. In addition, traditional statistical models have limitations in identifying factors associated with QoL over time, particularly concerning the physical, psychological, economic, spiritual, and social dimensions. This study aimed to identify patient-centered factors associated with QoL among patients with breast cancer using a machine learning (ML) algorithm to analyze data collected along different survivorship trajectories. The study used 2 data sets. The first data set was the cross-sectional survey data from the Breast Cancer Information Grand Round for Survivorship (BIG-S) study, which recruited consecutive breast cancer survivors who visited the outpatient breast cancer clinic at the Samsung Medical Center in Seoul, Korea, between 2018 and 2019. The second data set was the longitudinal cohort data from the Beauty Education for Distressed Breast Cancer (BEST) cohort study, which was conducted at 2 university-based cancer hospitals in Seoul, Korea, between 2011 and 2016. QoL was measured using European Organization for Research and Treatment of Cancer QoL Questionnaire Core 30 questionnaire. Feature importance was interpreted using Shapley Additive Explanations (SHAP). The final model was selected based on the highest mean area under the receiver operating characteristic curve (AUC). The analyses were performed using the Python 3.7 programming environment (Python Software Foundation). The study included 6265 breast cancer survivors in the training data set and 432 patients in the validation set. The mean age was 50.6 (SD 8.66) years and 46.8% (n=2004) had stage 1 cancer. In the training data set, 48.3% (n=3026) of survivors had poor QoL. The study developed ML models for QoL prediction based on 6 algorithms. Performance was good for all survival trajectories: overall (AUC 0.823), baseline (AUC 0.835), within 1 year (AUC 0.860), between 2 and 3 years (AUC 0.808), between 3 and 4 years (AUC 0.820), and between 4 and 5 years (AUC 0.826). Emotional and physical functions were the most important features before surgery and within 1 year after surgery, respectively. Fatigue was the most important feature between 1 and 4 years. Despite the survival period, hopefulness was the most influential feature on QoL. External validation of the models showed good performance with AUCs between 0.770 and 0.862. The study identified important factors associated with QoL among breast cancer survivors across different survival trajectories. Understanding the changing trends of these factors could help to intervene more precisely and timely, and potentially prevent or alleviate QoL-related issues for patients. The good performance of our ML models in both training and external validation sets suggests the potential use of this approach in identifying patient-centered factors and improving survivorship care.

Leakage and the reproducibility crisis in machine-learning-based science

Machine-learning (ML) methods have gained prominence in the quantitative sciences. However, there aremany known methodological pitfalls, including data leakage, in ML-based science. We systematically investigate reproducibility issues in ML-based science. Through a survey of literature in fields that haveadopted ML methods, we find 17 fields where leakage has been found, collectively affecting 294 papersand, in some cases, leading to wildly overoptimistic conclusions. Based on our survey, we introduce a detailed taxonomy of eight types of leakage, ranging from textbook errors to open research problems. We propose that researchers test for each type of leakage by filling out model info sheets, which weintroduce. Finally, we conduct a reproducibility study of civil war prediction, where complex ML models are believed to vastly outperform traditional statistical models such as logistic regression (LR). When theerrors are corrected, complex ML models do not perform substantively better than decades-old LR models.