Model Selection Research Articles

A deep learning method for comparing Bayesian hierarchical models.

Bayesian model comparison (BMC) offers a principled approach to assessing the relative merits of competing computational models and propagating uncertainty into model selection decisions. However, BMC is often intractable for the popular class of hierarchical models due to their high-dimensional nested parameter structure. To address this intractability, we propose a deep learning method for performing BMC on any set of hierarchical models which can be instantiated as probabilistic programs. Since our method enables amortized inference, it allows efficient re-estimation of posterior model probabilities and fast performance validation prior to any real-data application. In a series of extensive validation studies, we benchmark the performance of our method against the state-of-the-art bridge sampling method and demonstrate excellent amortized inference across all BMC settings. We then showcase our method by comparing four hierarchical evidence accumulation models that have previously been deemed intractable for BMC due to partly implicit likelihoods. Additionally, we demonstrate how transfer learning can be leveraged to enhance training efficiency. We provide reproducible code for all analyses and an open-source implementation of our method. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Reservoir temperature prediction based on characterization of water chemistry data—case study of western Anatolia, Turkey

Reservoir temperature estimation is crucial for geothermal studies, but traditional methods are complex and uncertain. To address this, we collected 83 sets of water chemistry and reservoir temperature data and applied four machine learning algorithms. These models considered various input factors and underwent data preprocessing steps like null value imputation, normalization, and Pearson coefficient calculation. Cross-validation addressed data volume issues, and performance metrics were used for model evaluation. The results revealed that our machine learning models outperformed traditional fluid geothermometers. All machine learning models surpassed traditional methods. The XGBoost model, based on the F-3 combination, demonstrated the best prediction accuracy with an R2 of 0.9732, while the Bayesian ridge regression model using the F-4 combination had the lowest performance with an R2 of 0.8302. This study highlights the potential of machine learning for accurate reservoir temperature prediction, offering geothermal professionals a reliable tool for model selection and advancing our understanding of geothermal resources.

Continuous Glucose, Insulin and Lifestyle Data Augmentation in Artificial Pancreas Using Adaptive Generative and Discriminative Models.

Artificial pancreas requires data from multiple sources for accurate insulin dose estimation. These include data from continuous glucose sensors, past insulin dosage information, meal quantity and time and physical activity data. The effectiveness of closed-loop diabetes management systems might be hampered by the absence of these data caused by device error or lack of compliance by patients. In this study, we demonstrate the effect of output sequence length-driven generative and discriminative model selection in high quality data generation and augmentation. This novel generative adversarial network (GAN) based architecture automatically selects the generator and discriminator architecture based on the desired output sequence length. The proposed model is able to generate glucose, physical activity, meal information data for individual patients. The discriminative scores for Ohio T1DM (2018) dataset were 0.17 ±0.03 (Inputs: CGM, CHO, Insulin) and 0.15 ±0.02 (Inputs: CGM, CHO, Insulin, Heart Rate, Steps) and for Ohio T1D (2020) dataset was 0.16 ±0.02 (Inputs: CGM, CHO, Insulin) and 0.15 ±0.02 (Inputs: CGM, CHO, Insulin, acceleration). A mixture of generated and real data was used to test predictive scores for glucose forecasting models. The best RMSE and MARD achieved for OhioT1DM patients were 17.19 ±3.22 and 7.14 ±1.76 for PH=30min with CGM, CHO, Insulin, heartrate and steps as inputs. Similarly, the RMSE and MARD for real+synthetic data were 15.63 ±2.57 and 5.86 ±1.69 respectively. Compared to existing generative models, we demonstrate that sequence length based architecture selection leads to better synthetic data generation for multiple output sequences (CGM, CHO, Insulin) and forecasting accuracy.

78 Estimates of genetic parameters for weight, wool, and carcass traits of Rambouillet sheep

Abstract The Rambouillet sheep breed is the foundation to most western range sheep in the U.S. Through the National Sheep Improvement Program (NSIP), Rambouillet breeders seek to improve economically relevant traits. However, the parameters used in the NSIP genetic evaluation were last estimated two decades ago. Our aim was to re-estimate genetic parameters for body weight (BW), carcass, and wool traits. Data were provided by NSIP, which included BW at birth (BWT; n = 28,817), weaning [WWT; n = 5,633; age: 68 (SD 16) day], early post-weaning [PWTE; n = 23,303; 128 (SD 22) d], late post-weaning [PWTL; n = 7,689; 201 (SD 45) d] and as yearlings [YWT; n = 5,830; 375 (SD 33) d]. Yearling ultrasound scans of 12th rib backfat depth (YCF; n = 2,447) and eye muscle depth (YEMD; n = 2,457) were also available. Yearling wool traits were fiber curvature (YFC; n = 6,009), fiber diameter (YFD; n = 9,880), fiber diameter coefficient of variation (YFDCV; n = 8,803), greasy fleece weight (YGFW; n = 11,859) and staple length (YSL; n = 11,270). The pedigree consisted of 36,297 individuals. Animals with phenotypic records were progeny of 333 sires and 1,727 dams. A succession of univariate animal models was fitted. The most comprehensive model included direct and maternal additive genetic effects, their genetic covariance, and maternal permanent environmental and residual effects. Phenotypes were pre-adjusted for birth-rearing type category and, as covariates, dam age and age at recording. Contemporary group—a concatenation of flock, year, season, management group, sex, and, excluding BWT, recording date—was the only fixed effect fitted in the model. Model selection was based on the log-likelihood ratio test, and Akaike and Bayesian information criteria. The data structure, namely dams and their daughters and granddaughters contributing data, allowed reliable testing of the maternal effect terms. The direct-maternal genetic covariance did not explain variation for any trait (P > 0.18). The most parsimonious model for BW traits included direct and maternal additive genetic, and maternal permanent environmental, effects. For carcass and wool traits, only direct additive genetic effects were included in the final model. Such was unsurprising, as maternal effects often diminish as animals age. For the final models fitted (Table 1), direct heritabilities (± SE) for body weights ranged from 0.14 ± 0.01 (BWT) to 0.38 ± 0.04 (PWTL); the maternal heritabilities ranged from 0.04 ± 0.02 (PWTL; YWT) to 0.13 ± 0.01 (BWT). For carcass traits, direct heritabilities were 0.13 ± 0.04 (YCF) and 0.31 ± 0.05 (YEMD). For wool traits, they ranged from 0.37 ± 0.02 (YSL) to 0.58 ± 0.02 (YFD). These re-estimated parameter values are consistent with those used by NSIP, with the exception of PWTL (greater) and YSL (lower). With genomic data accumulating in Rambouillet, further refinements in parameter estimation are planned.

Local Watershed Properties Cannot Explain Divergent Dynamics of Pacific Herring in an Urbanizing Estuary

AbstractPacific herring (Clupea pallasii) is a foundational species in Puget Sound (Washington State, U.S.A.) and is subject to many anthropogenic threats. We assessed the overall status of the Puget Sound Pacific herring sub-stock complex and asked whether watersheds with less urban or agricultural land cover, less impervious surface, and lower human density were associated with better stock status. To this end, we developed multiple metrics of sub-stock population status; characterized watershed properties with respect to land use/land cover, percent impervious surfaces, and human density; and used statistical model selection to evaluate the weight of evidence in support of our hypotheses. Overall, the status of sub-stocks was poor; metrics for most sub-stocks indicate a decline from 1996–2021. However, the status metrics of sub-stocks were not related to recent (2016) watershed characteristics or the rate of change in watershed characteristics from the mid-1990s to 2016. While the cumulative effects of local human land use throughout Puget Sound may be contributing to the deterioration of spawning biomass, these results also suggest that other drivers that operate at larger scales (e.g., predation, disease, climate) are likely important.

Estimating the risk of SARS-CoV-2 infection in New Zealand border arrivals

BackgroundTravel restrictions and border controls were used extensively during the COVID-19 pandemic. However, the processes for making robust evidence-based risk assessments of source countries to inform border control policies was in many cases very limited.MethodsBetween April 2020 and February 2022, all international arrivals to New Zealand were required to spend 14 days in government-managed quarantine facilities and were tested at least twice. The infection rates among arrivals in the years 2020, 2021 and 2022 were respectively 6.3, 9.4 and 90.0 cases per thousand arrivals (487, 1064 and 1496 cases). Test results for all arrivals were linked with travel history, providing a large and comprehensive dataset on the number of SARS-CoV-2-positive and negative travellers from different countries over time. We developed a statistical model to predict the country-level infection risk based on infection rates among recent arrivals and reported cases in the country of origin. The model incorporates a country-level random effect to allow for the differences between the infection risk of the population of each country and that of travellers to New Zealand. A time dependent auto-regressive component of the model allows for short term correlation in infection rates.ResultsA model selection and checking exercise found that the model was robust and reliable for forecasting arrival risk for 2 weeks ahead. We used the model to forecast the number of infected arrivals in future weeks and categorised countries according to their risk level. The model was implemented in R and was used by the New Zealand Ministry of Health to help inform border control policy during 2021.ConclusionsA robust and practical forecasting tool was developed for forecasting infection risk among arriving passengers during a period of controlled borders during the COVID-19 pandemic. The model uses historical infection rates among arrivals and current infection rates in the source country to make separate risk predictions for arrivals from each country.

The Implementation of the Case Method with Blooket Media to Enhance Student Learning Outcomes in Radiochemical Material

Research related to the implementation of learning models to improve the quality of learning continues to be carried out in line with changes in the applicable curriculum. The selection of the right learning model with the help of fun game media is the basis for this research, so that learning becomes more meaningful and not boring. This is a quasi-experimental study using a "nonrandomized pretest-posttest control group design."  Before the t-test, a prerequisite test was carried out in the form of a normality test (using the Liliefors test) and a homogeneity test (using the Levene test). In addition to hypothesis testing, this study also calculated the score increase (N-Gain) from the pretest and posttest data. The t count value is 2.88, while the t table value at = 0.05 with dk = 51 is 1.67. The hypothesis is supported because the t count value is greater than the t table value (2.88 > 1.67). In addition to the t count and t table values, the r2 = 0.1403 value was examined.  This coefficient value of 14.03% indicates the level of effect of employing the case method with blooket media on increasing student learning outcomes in the Radiochemistry subject.

Preferred Learning Model in Studying English as a Foreign Language During Post-Pandemic Period: Blended Learning, Online Learning, or Face to Face?

Objective: Before and During the Covid-19 pandemic, students in Indonesia experienced face to face learning, online learning, and blended learning. Since the students have experienced those three learning models, this study was conducted to investigate the learning model that is preferred by the students during post-pandemic period and identify the reasons why they chose a particular learning model in learning English as a foreign language context.   Method: To achieve that research objective, this study was conducted by following a mixed-methods study method, the sequential explanatory design. The researchers started this study by collecting the quantitative data to identify the learning models preferred by the university students in Bali when they have an English class during post-pandemic period. Then, it was followed by collecting qualitative data to explain why the students preferred a certain learning model for the English class. The respondents of the study were 100 lecturers and 200 university students. The researchers collected the study data through questionnaires and interviews. The questionnaires were delivered using Google Forms, and the interviews via Google Meet.   Results and Discussion: The study reveals that most lecturers and students preferred the blended learning model to teach and learn English. In general, the selection of the learning model preferred by the lecturers and students was influenced by educational and health factors. Therefore, it can be concluded that the students’ preference on selecting a learning model is not influence by the educational reasons only.   Research Implications: Face to face instruction is crucial in an education process, because a real teacher-student interaction will happen through face to face instruction. Thus, blended learning is more preferable by the students than e-learning.   Originality/Value: Previous studies on blended in Indonesia were conducted to reveal the effectiveness of blended learning education for the broader use of blended learning in modern learning. Only a few studies were conducted to reveal their effectiveness in limited face-to-face learning during the post-pandemic of Covid-19. Therefore, this study mainly focused on investigating the effectiveness of blended learning during the implementation of limited-face to-face learning in Indonesia during the post-pandemic of the covid-19 pandemic. This study was necessary because the findings can guide educators to develop a more effective learning environment.

Novel autoantibodies help diagnose anti-SSA antibody negative Sjögren disease and predict abnormal labial salivary gland pathology

ObjectivesSjögren disease (SjD) diagnosis often requires either positive anti-SSA antibodies or a labial salivary gland biopsy with a positive focus score (FS). One-third of patients with SjD lack anti-SSA antibodies...

Effect of the early diastolic blood pressure response to the head-up tilt test on the recurrence of benign paroxysmal positional vertigo.

Otolith organ acts complementarily with the autonomic nervous system to maintain blood pressure. However, the effect of blood pressure variability in the autonomic nervous system on otolith organ has not yet been determined. This study aimed to verify the hypothesis that blood pressure variability in the autonomic nervous system affects the recurrence of benign paroxysmal positional vertigo (BPPV), which is the most common disease of the vestibular organs, by using the head-up tilt test (HUTT). This study included 432 patients diagnosed with idiopathic BPPV. The follow-up period for all patients was 12 months. Age, sex, hypertension, diabetes and recurrence were analyzed. The HUTT parameters were divided into a group of patients whose average diastolic blood pressure increased in the upright position compared to supine position during the HUTT (DBP1) and a group of patients whose average diastolic blood pressure decreased in the upright position compared to supine position during the HUTT (DBP2). Model selection, general loglinear analysis, and logit loglinear analysis were performed using a hierarchically progressing loglinear analysis. In summary, the group with increased average diastolic blood pressure (DBP1) showed a higher tendency for BPPV recurrence compared to the group with decreased diastolic blood pressure (DBP2) in the upright position during the HUTT, although the difference was not statistically significant (p = 0.080). However, in males, the DBP1 group demonstrated a significantly higher recurrence rate of BPPV than the DBP2 group during the HUTT (95% CI, -20.021 to -16.200; p < 0.001). It is presumed that poor autonomic nervous system response through vestibulosympathetic reflex maintains elevated diastolic blood pressure in the upright position during the HUTT. This variability is assumed to affect the recurrence of BPPV.

BITCOIN PRICE PREDICTION BY USING ARIMA

Cryptocurrency markets have emerged as a dynamic and intriguing domain, with Bitcoin at the forefront, captivating the attention of investors, researchers, and enthusiasts alike. The volatile nature of Bitcoin prices presents both opportunities and challenges for market participants seeking to understand and anticipate its movements. In this study, we delve into the realm of time series analysis to explore the feasibility of predicting The research journey begins with meticulous data preprocessing steps to ensure the quality and integrity of the input data. Leveraging Python libraries such as pandas and NumPy, we cleanse and format the historical Bitcoin price data, laying the foundation for subsequent analysis. Key preprocessing tasks include handling missing values, normalization, and addressing any anomalies or outliers that may distort the underlying patterns. With the data prepared, our attention turns to assessing the stationarity of the Bitcoin price time series—a fundamental prerequisite for applying classical time series models. Through visual inspection and statistical tests such as the Augmented Dickey-Fuller (ADF) test, we ascertain the presence of trends or seasonality that could influence the modelling process. To mitigate such effects, we employ techniques such as differencing and transformations, including the Box-Cox transformation, to stabilize the variance of the data. Armed with a stationary time series, we embark on the core of our analysis: modelling Bitcoin prices using Autoregressive Integrated Moving Average (ARIMA) and Seasonal Autoregressive Integrated Moving Average with Exogenous Variables (SARIMAX) models. These models, renowned for their versatility and effectiveness in capturing temporal dependencies, offer a sophisticated framework for forecasting time series data. Guided by the principles of parsimony and model selection criteria such as the Akaike Information Criterion (AIC), we systematically explore the parameter space to identify the optimal specifications for our models. The efficacy of the chosen models is rigorously evaluated through diagnostic checks, encompassing residual analysis, model fit statistics, and out-of-sample validation. Insights gleaned from these assessments inform our confidence in the models' predictive capabilities and guide our interpretation of the forecasted outcomes. Finally, armed with a validated model, we turn our gaze to the future, employing it to generate forecasts of Bitcoin prices for forthcoming periods. Visualizations juxtaposing predicted prices against observed values provide a compelling narrative of the model's performance and offer stakeholders valuable insights into potential market trends and dynamics. In summary, this research contributes to the burgeoning field of cryptocurrency analytics by showcasing the application of time-tested statistical methodologies to forecast Bitcoin prices. By leveraging the power of ARIMA and SARIMAX models, we illuminate the intricate patterns underlying Bitcoin's price dynamics, empowering market participants with actionable intelligence for informed decision-making in an ever-evolving landscape. Keywords Cryptocurrency Markets, Bitcoin Price Prediction, Time Series Analysis, Python Programming, Data Preprocessing, Pandas, NumPy, Stationarity Testing, Augmented Dickey-Fuller (ADF) Test, Box-Cox Transformation, ARIMA Model, SARIMAX Model, Model Selection, Akaike Information Criterion (AIC), Diagnostic Checks, Residual Analysis, Out-of-Sample Validation, Forecasting, Visualization, Market Trends, Decision-Making, Cryptocurrency Analytics

USER PERCEPTION TOWARDS PREDICTIVE ANALYTICS IN CREDIT RISK MANAGEMENT

This comprehensive review explores the dynamic landscape of predictive analytics in credit risk management within the banking sector. Anchored in a qualitative research design, the study synthesizes existing literature and real-world case studies to provide a multifaceted understanding of predictive analytics' role in modern banking. The review identifies key trends, highlighting the integration of predictive analytics across diverse banking operations, the transition to advanced machine learning algorithms, the democratization of predictive analytics tools, and the growing emphasis on ethical and regulatory compliance. It underscores the effectiveness of predictive analytics, showcasing its ability to enhance risk assessment precision, decision-making agility, and overall banking performance. Comparative analyses reveal the varying performance of predictive models across contexts, emphasizing the importance of tailored model selection. However, challenges such as data quality, model interpretability, talent scarcity, ethical considerations, and implementation costs pose significant hurdles. Looking forward, predictive analytics promises to be an indispensable tool for mitigating credit risk in the banking sector, offering refined risk assessments, smarter decisions, and enhanced resilience. The insights from this review provide valuable guidance for banking professionals, regulators, and researchers navigating the evolving landscape of predictive analytics in banking.

Impact of Model Selection and Conformational Effects on the Descriptors for In Silico Screening Campaigns: A Case Study of Rh-Catalyzed Acrylate Hydrogenation

Impact of Model Selection and Conformational Effects on the Descriptors for In Silico Screening Campaigns: A Case Study of Rh-Catalyzed Acrylate Hydrogenation

Abstract PS04-04: Adverse effects (AEs) reported via Patient Reported Outcome Measures (PROMs): A longitudinal analysis of the IMPORT HIGH breast radiotherapy trial (CRUK/06/003)

Abstract Background IMPORT HIGH is a multicentre phase III randomised controlled trial showing efficacy and safety of appropriately-dosed simultaneous integrated boost (SIB) in terms of local control, patient and clinician reported AEs among women with early breast cancer (BC)[1]. The PROMs sub-study collected baseline demographics and quality of life over 5 years. This exploratory analysis details the patient experience over time. Methods Women with higher-than-average risk of local relapse (≥18 years, breast conserving surgery, T1-3, pN0-pN3a, M0 invasive BC) were eligible. Randomisation (1:1:1) was between 40Gy whole breast and 16Gy sequential boost radiotherapy (control) and, 36Gy whole breast, 40Gy partial breast with either 48Gy SIB (test group 1), or 53Gy SIB (test group 2) radiotherapy. AEs from EORTC QLQ-BR23[2]; Body Image Scale (BIS)[3]; and protocol-specific (PS) items used in previous breast radiotherapy trials[4] were reported by patients at baseline (pre-randomisation), 6 months, and years 1, 3, 5. 24 AEs[4] were included in the analysis, dichotomised as none/mild versus moderate/marked. Anxiety and depression were measured over 5 years using Hospital Anxiety and Depression Scale. Surgical deficit was scored from baseline photographs. Multivariable generalized estimating equation models were fitted for each AE. Potential baseline predictors included those used previously[4] plus ethnicity, smoking, deprivation, employment, cardiovascular disease history, post-operative infection, haematoma, post-surgery complication. Backwards selection and Quasi-likelihood criterion were used for model selection. Results Between 2009-2013, 1338 patients were offered the PROMs sub-study and 1078 (81%) participated. Median age was 48.8 years (IQR 44.0-54.9). 17% reported ‘Borderline’ or 14% ‘Case’ anxiety at baseline. 11% reported ‘Borderline’ or 7% ‘Case’ depression at baseline. 747/1015 (74%), 686/986 (70%), 604/916 (66%), 532/845 (63%), 461/709 (65%) of participants reported at least 1 moderate/marked AE at baseline, 6 months, 1, 3 and 5 years. Number of moderate/marked AEs reported per person reduced over time after year 1 (p &amp;lt; 0.005). BIS reported moderate/marked AEs (except dissatisfaction with scar) decreased, whereas individual items from BR23 and PS items showed mixed patterns of reporting (increased, decreased, stayed the same) over time (Table 1). From multivariable analysis, ‘Borderline’ or ‘Case’ depression or anxiety (time-varying) was associated with increased reports for almost all AEs. Chemotherapy and younger age were associated with increased reports of moderate/marked AEs for items in BIS, and large surgical deficit predicted more breast symptoms. Conclusion Change of overall breast appearance was the most reported moderate/marked AE that persisted over time.Although patients reported fewer moderate/marked AEs over time, trends in individual moderate/marked AEs showed mixed patterns. This study identified factors associated with increased risks of patient-reported moderate/marked AEs. Early identification of patients more likely to report AEs could facilitate implementation of targeted support.

Estimating the population size of persons contending with homelessness using electronic health records

Abstract The majority of attempts to enumerate the homeless population rely on point-in-time or shelter counts, which can be costly and inaccurate. As an alternative, we use electronic health records from the Vancouver Island Health Authority, British Columbia, Canada from 2013 to 2022 to identify adults contending with homelessness based on their self-reported housing status. We estimate the annual population size of this population using a flexible open-population capture–recapture model that takes into account (1) the age and gender structure of the population, including aging across detection occasions, (2) annual recruitment into the population, (3) behavioural-response, and (4) apparent survival in the population, including emigration and incorporating known deaths. With this model, we demonstrate how to perform model selection for the inclusion of covariates. We then compare our estimates of annual population size with reported point-in-time counts of homeless populations on Vancouver Island over the same time period, and find that using data extracts from electronic health records gives comparable estimates. We find similarly comparable results using only a subset of interaction data, when using only ER interactions, suggesting that even if cross-continuum data is not available, reasonable estimates of population size can still be found using our method.

Estimating a regression function in exponential families by model selection

Estimating a regression function in exponential families by model selection

THE INFLUENCE OF RESISTANCE TRAINING VARIABLES ON MUSCLE STRENGTH: A SYSTEMATIC REVIEW AND META-ANALYSIS

BACKGROUND Development of muscle strength through resistance training (RT) is associated with health and performance benefits. Prescribing resistance training utilising six different variables (sets × repetitions × exercises × intensity × frequency × duration) is complex to optimise outcomes. It is unclear if, the prescription of these variables to derive a composite variable, RT dosage, influences the development of muscle strength. OBJECTIVES: This study aimed to investigate if variations in RT dosage influence muscle strength. METHODS CINAHL, MEDLINE, and SPORTDiscus were systematically searched; adults performing RT were compared against non-exercise controls where muscle strength was assessed. Only randomised controlled trials assessing chest and quadriceps muscle strength were considered. The six RT variables were extracted from eligible studies and were multiplied to generate RT dosage. Quadratic non-linear regressions were performed on pooled muscle strength data. Variable importance and model selection were performed to explore different linear combinations of RT variables to determine the most effective variable in developing muscle strength. RESULTS 207 articles were eligible that enabled the calculation of RT dosage. Meta-analyses confirmed a significant effect of RT on muscle strength (SMD ≥ 1.3, CI = 1.1-1.6, p &amp;lt; 0.001). Quadratic non-linear regression models for the chest and quadriceps strength indicated that RT dosage as a continuous variable showed a significant dose-response for muscle strength (SMD = &amp;gt; 1.5, CI = 0.01-3.1, P &amp;lt; 0.05). The individual variables of most importance to develop muscle strength appear to be volume and duration (p &amp;lt; 0.03). CONCLUSIONS As expected, RT significantly influences muscle strength and as RT dosage increases, so too does muscle strength; however, only up to a point (∼1.5 million arbitrary units) before further increases in RT dosage don’t appear to further develop muscle strength. Prescription of RT volume and duration should be prioritised for developing muscle strength.

Empirical likelihood ratio tests for non-nested model selection based on predictive losses

Empirical likelihood ratio tests for non-nested model selection based on predictive losses

Predicting types of human-related maritime accidents with explanations using selective ensemble learning and SHAP method

Maritime accidents frequently lead to severe property damage and casualties, and an accurate and reliable risk prediction model is necessary to help maritime stakeholders assess the current risk situation. Therefore, the present study proposes a hybrid methodology to develop an explainable prediction model for maritime accident types. Based on the advantages of selective ensemble learning method, this study pioneers to introduce a two-stage model selection method, aiming to enhance the predictive accuracy and stability of the model. Then, SHAP (Shapley Additive Explanations) method is integrated to identify effective mapping associations of seafarers’ unsafe acts and their risk factors with the prediction results. The results demonstrate that the model developed achieves good prediction performance with an accuracy of 87.50% and an F1-score of 84.98%, which benefits stakeholders in assessing the type of maritime accident in advance, so as to make proactive intervention measures.

MLCompare: Platform to Compare Machine Learning Models

Abstract: The field of machine learning (ML) continues to advance, the demand for efficient and effective methods for model development and selection becomes increasingly pressing. Automated Machine Learning (AutoML) platforms have emerged as powerful tools to address this need by automating the process of model selection, hyperparameter tuning, and feature engineering. By integrating a wide array of ML techniques and leveraging cloud computing resources, the platform enables researchers and practitioners to efficiently explore and compare the performance of various algorithms across diverse datasets. Key features include a user-friendly interface, a robust evaluation framework encompassing traditional metrics and advanced techniques, and parallel processing capabilities. Through experiments on benchmark datasets from multiple domains, the platform demonstrates its ability to streamline algorithm selection and provide valuable insights into algorithm performance. This contribution advances the field of AutoML by facilitating informed decision-making and accelerating innovation in ML applications.