Class Cluster Research Articles

An interpretable deep learning model for detecting BRCA pathogenic variants of breast cancer from hematoxylin and eosin-stained pathological images

Background Determining the status of breast cancer susceptibility genes (BRCA) is crucial for guiding breast cancer treatment. Nevertheless, the need for BRCA genetic testing among breast cancer patients remains unmet due to high costs and limited resources. This study aimed to develop a Bi-directional Self-Attention Multiple Instance Learning (BiAMIL) algorithm to detect BRCA status from hematoxylin and eosin (H&E) pathological images. Methods A total of 319 histopathological slides from 254 breast cancer patients were included, comprising two dependent cohorts. Following image pre-processing, 633,484 tumor tiles from the training dataset were employed to train the self-developed deep-learning model. The performance of the network was evaluated in the internal and external test sets. Results BiAMIL achieved AUC values of 0.819 (95% CI [0.673–0.965]) in the internal test set, and 0.817 (95% CI [0.712–0.923]) in the external test set. To explore the relationship between BRCA status and interpretable morphological features in pathological images, we utilized Class Activation Mapping (CAM) technique and cluster analysis to investigate the connections between BRCA gene mutation status and tissue and cell features. Significantly, we observed that tumor-infiltrating lymphocytes and the morphological characteristics of tumor cells appeared to be potential features associated with BRCA status. Conclusions An interpretable deep neural network model based on the attention mechanism was developed to predict the BRCA status in breast cancer. Keywords: Breast cancer, BRCA, deep learning, self-attention, interpretability.

Exploring the Differences in Types of Commodities and Delivery Methods among Online Shoppers during the Early Stage of the COVID-19 Pandemic

During the early phases of the COVID-19 pandemic, the temporary limitations to in-store shopping during the lockdown period generated a significant increase in online shopping. During that period, the total spending amount and the number of online purchases increased across the whole population, although with rather heterogeneous patterns among various groups of consumers. This research examines the factors that can help explain the differences between users and non-users of online shopping in the U.S.A. during April and May 2020 to help understand the barriers to home-based shopping accessibility. The study uses data collected in the first survey wave of a longitudinal mobility study to analyze individual commodities and delivery options chosen for goods purchased online. Binomial logistic regression and Poisson regression models reveal differences in adopting e-shopping across various delivery channels, the frequency of purchase, and shipping used across commodities. In addition, a latent class cluster analysis identifies three heterogeneous groups of shoppers whose members differ mainly in income, age, education, neighborhood type, and technology attitudes. The results also highlight several differences in usage across crowdshipping markets and novel delivery options, such as in-store or locker pickup. The results also show variation in home-based accessibility across numerous factors during the pandemic.

Who is ready to live a car-independent lifestyle? A latent class cluster analysis of attitudes towards car ownership and usage

An excessive use of private cars for transportation has multiple negative effects on our society. It is therefore paramount to identify the underlying factors driving car usage among different segments of travellers. In this study, we aim to identify and characterise traveller segments in terms of their car-related attitudes and how different sociodemographic attributes, travel behaviour characteristics (such as using cars as the primary mode of transportation), and the accessibility of their place of residence vary amongst different population segments. Through Confirmatory Factor Analysis and Latent Profile Analysis we identify five different classes, namely car-detractors, −hesitants, −positives, −friends, and −lovers. More favourable car attitudes are associated with higher car ownership and access but, interestingly, there are no significant differences in terms of travel frequency and trip length distribution. This suggests that attitudinal differences relate mostly to modal preference rather than to the underlying travel demand patterns and the associated activities. We also see that car-less individuals are more likely to be car detractors, suggesting that the experiencing barriers for owning a car may contribute to developing more negative attitudes towards cars. Our results shed light on the different influencing factors and characteristics of individuals that consider, or not, a car-centred mobility.

Burn Injury Severity in Adults: Proposed Definitions Based on the National Burn Research Dataset.

Previous iterations of burn severity (mild, moderate, and severe) were not data-driven and were outdated. Clustering analyses have gained popularity for identifying homogenous subgroups from heterogeneous medical conditions, such as asthma, sepsis, and lung disease. There is no consensus in burn literature regarding what constitutes massive burns. The current classification includes a 20% total body surface area (TBSA) burn and a 95% TBSA burn as severe. Latent class and hierarchical clustering analyses were applied to the American Burn Association National Burn Research Dataset. Cluster variables included length of stay, length of stay, intensive care unit length of, number and type of procedures, and number and type of complications. Non-clustering variables were evaluated after clustering, including burned TBSA, inhalation injury, mortality, discharge disposition, age, sex, and race. Latent class analysis suggested three clusters. Hierarchical clustering analysis was applied to the most severe latent class, creating four total burn severity groups. In total, 112,297 patients were included in the final analysis. The mean TBSA burned for each class is 4.26±4.91 for minor, 8.07±8.39 for moderate, 22.76±17.31 for severe and 36.72±21.61 for massive. The age and sex proportions were similar among all clusters. The clustering variables steadily increased for each severity cluster. Mortality was the highest in the massive cluster (18.2%). Data informed categories of burn severity were formed using clustering analyses, which will be helpful for triage, data-benchmarking, and class-specific research.

Deep Learning-Driven Prediction of Mechanical Properties of 316L Stainless Steel Metallographic by Laser Powder Bed Fusion.

This study developed a new metallography-property relationship neural network (MPR-Net) to predict the relationship between the microstructure and mechanical properties of 316L stainless steel built by laser powder bed fusion (LPBF). The accuracy R2 of MPR-Net was 0.96 and 0.91 for tensile strength and Vickers hardness predictions, respectively, based on optical metallurgy images. Feature visualisation methods, such as gradient-weighted class activation mapping (Grad-CAM) and clustering, were employed to interpret the abstract features within the MPR-Net, providing insights into the molten pool morphology and grain formation mechanisms during the LPBF process. Experimental results showed that the optimal process parameters-190 W laser power and 700 mm/s scanning speed-yielded a maximum tensile strength of 762.83 MPa and a Vickers hardness of 253.07 HV0.2 with nearly full densification (99.97%). The study marks the first application of a convolutional neural network (MPR-Net) to predict the mechanical properties of 316L stainless steel samples manufactured through laser powder bed fusion (LPBF) based on metallography. It innovatively employs techniques such as gradient-weighted class activation mapping (Grad-CAM), spatial coherence testing, and clustering to provide deeper insights into the workings of the machine learning model, enhancing the interpretability of complex neural network decisions in material science.

Investigation of the effects of polyethylene microplastics at environmentally relevant concentrations on the plant-soil-microbiota system: A two-year field trial

Microplastics are a potential threat to agricultural sustainability. However, the effects of microplastics at environmentally relevant concentrations on the plant-soil-microbiota system in realistic field conditions are largely unknown. Herein, we conducted a two-year field trial to study the effects of polyethylene (PE) microplastics at 0, 100, and 600 mg/kg on crop growth, soil properties, and the composition and function of microbial communities in a farmland with rice-wheat rotation. PE did not affect wheat growth but it increased the rice grain weight by 42.5 % at 600 mg/kg, and enhanced rice height by 35.4 % and 30.2 % at 100 and 600 mg/kg, respectively. The presence of PE significantly decreased soil available phosphorus during the wheat season, while it reduced soil total nitrogen, NH4+-N and available phosphorus during the rice season. There were five and sixteen bacterial orders identified changed by PE in wheat and rice soils, respectively. Specifically, PE at different concentrations differentially altered the abundances of sulfate-reducing bacteria Thermodesulfovibrionia, Thermoactinomycetales and Syntrophobacterales, and further modified soil sulfate respiration in wheat soils. During the rice season, PE (100 mg/kg) increased the abundance of Xanthomonadales by 98.0 % and enriched the functional groups of intracellular parasites, while PE (600 mg/kg) inhibited twelve cluster of orthologous group function classes and disturbed bacterial metabolism. This study suggests that PE exhibits a greater impact on the plant-soil-microbiota system during the rice season compared to the previous year's wheat season, highlighting the importance of crop type and cultivation practices in determining the environmental risks of microplastics in agroecosystems.

Identifying Population Segments by Differing Levels of COVID-19 Vaccine Confidence and Evaluating Subsequent Uptake of COVID-19 Prevention Behaviors: Web-Based, Longitudinal, Probability-Based Panel Survey.

The COVID-19 pandemic prompted the launch of the US Department of Health and Human Services' COVID-19 Public Education Campaign to boost vaccine confidence and uptake among adults, as vaccines are key to preventing severe illness and death. Past segmentation research relevant to COVID-19 behavior has found important differences in attitudes, sociodemographics, and subsequent COVID-19 prevention behaviors across population segments. This study extends prior work by incorporating a more comprehensive set of attitudes, behaviors, and sociodemographic variables to identify population segments by differing levels of COVID-19 vaccine confidence and evaluate differences in their subsequent uptake of COVID-19 prevention behaviors. Data were obtained from 5 waves (January 2021 to June 2022) of a web-based longitudinal, probability-based panel survey of US adults (N=4398) administered in English and in Spanish. Participants were recruited from NORC at the University of Chicago's national AmeriSpeak panel and were invited to participate across multiple waves. Latent class cluster analysis estimated segments of respondents based on over 40 COVID-19 attitudes, beliefs, behaviors, and sociodemographics as reported in wave 1. Survey-weighted cross-tabulations and bivariate regression analyses assessed differences in COVID-19 vaccine uptake, booster uptake, mask use, and social distancing in all segments across all 5 survey waves. A total of 6 segments (hardline nonintenders, prevention-compliant nonintenders, burned-out waiters, anxious waiters, skeptical confidents, and ready confidents) were identified, which differed by their COVID-19 vaccine confidence, prevention-related attitudes and behaviors, and sociodemographics. Cross-tabulations and regression results indicated significant segment membership differences in COVID-19 vaccine and booster timing, mask use, and social distancing. Results from survey-weighted cross-tabulations comparing COVID-19 vaccine and booster uptake across segments indicate statistically significant differences in these outcomes across the 6 segments (P<.001). Results were statistically significant for each segment (P<.01 for booster uptake among burned-out waiters; P<.001 for all other coefficients), indicating that, on average, respondents in segments with lower intentions to vaccinate reported later receipt of COVID-19 vaccines and boosters relative to the timing of vaccine and booster uptake among ready confidents. Results extend previous research by showing that initial beliefs and behaviors relevant to COVID-19 vaccination, mask use, and social distancing are important for understanding differences in subsequent compliance with recommended COVID-19 prevention measures. Specifically, we found that across respondent segments, the probability of vaccine and booster uptake corresponded with both COVID-19 vaccine confidence and mask use and social distancing compliance; more compliant segments were more likely to get vaccinated or boosted than less compliant segments given similar levels of vaccine confidence. These findings help identify appropriate audiences for campaigns. Results highlight the use of a comprehensive list of attitudes, behaviors, and other individual-level characteristics that can serve as a basis for future segmentation efforts relevant to COVID-19 and other infectious diseases.

Unfolding multilevel agglomerative strategies for SVM classification: a case study in discriminating spectrally similar land covers

ABSTRACT In remote sensing applications, image classification algorithms normally require parameter optimization strategies to adapt to the complexities of the data and determine the model’s parameters for optimal performance. However, a unique set of hyper-parameters is generally chosen, without acknowledging the presence of subsets of classes that exhibit different levels of spectral similarity. A model used to distinguish between well-separable classes might not be optimal for delineating other classes. This study explores a classification method that extends the standard One-Against-One (OAO) and One-Against-All (OAA) strategies for Support Vector Machines (SVMs), based on agglomerative hierarchical clustering of spectrally-similar classes. The main objective is to provide additional classification scenarios where more detailed per-class analyses can be performed, rather than simple comparisons of global accuracies, particularly for hard-to-distinguish classes. Clusters of classes closely located in the feature space are initially classified using an OAA strategy, while each cluster undergoes further subdivision using an OAO approach. This strategy aims to identify the most effective sequence for incorporating subgroups of classes into the tuning process. The experimental results indicate that the proposed method can be efficiently applied to classify remote sensing data, producing varying levels of enhancement in per-class accuracies at each stage of the agglomerative process.

Latent Class Analysis Reveals, in patient profiles, COVID-19–related better prognosis by calcifediol treatment than glucocorticoids

Calcifediol and glucocorticoids have been repositioned for the treatment of COVID-19 and may reduce severity, the need for intensive care unit admission and death. Objectiveto identify class or profiles of patients hospitalized and treated with COVID-19 pneumonia using latent class clustering methods to assess the clinical and prognostic relevance of the resulting patients’ profiles. Poor prognosis was defined as death or need for ICU admission, good prognosis, the opposite. With special interest in differential responses to calcifediol. SettingReina Sofia University Hospital, Córdoba Spain. PatientsRetrospective observational cohort study of patients admitted for COVID-19. ClinicalTrials.gov public database (NCT05819918). Inclusion criteria: (i) Age ≥ 18 and ≤ 90 years, (ii) Pneumonia characterized by the presence of infiltrates on chest X-ray or CT scan, (iii) SARS-CoV-2 infection, confirmed, and (iv) CURB Scale 65 >1. DesignLatent class analysis, for obtaining homogeneous clusters, without specifying a priori the belonging group, and selecting the optimal number of clusters by minimizing information criteria. Evaluating the differences between groups for each variable by means of chi-square, Fisher's exact test and Kruskal-Wallis test. Results707 patients hospitalized from 10 March 2020 until 4 March 2022 were included. For the treatment variable, differences were found between class 3 (60 % treated with calcifediol only) and classes 1 (less than 1 % calcifediol only vs. 82 % treated with both), 2 (less than 1 % calcifediol only vs. 82 % treated with both) and 4 (1 % calcifediol only vs. 84 % treated with both). Class 3, (60 % with calcifediol), had a significantly better prognosis compared to patients treated with glucocorticoids alone (OR: 15.2, 95 % CI: [3.73–142], p<0.001) or no treatment (OR: 7.38, 95 % CI: [2.63–30.2], p<0.001). Conclusionsour real-life study shows that calcifediol treatment significantly reduces the need for ICU admission and improved prognosis in patients hospitalized for COVID-19 pneumonia, especially in the profile of patients receiving it without glucocorticoids.

Quantifying the heterogeneity impact of risk factors on regional bicycle crash frequency: A hybrid approach of clustering and random parameter model

The existence of internal and external heterogeneity has been established by numerous studies across various fields, including transportation and safety analysis. The findings from these studies underscore the complexity of crash data and the multifaceted nature of risk factors involved in accidents. However, most studies consider the effects of unobserved heterogeneity from one perspective −- either within clusters (internal) or between clusters (external) −- and do not investigate the biases from both simultaneously on crash frequency analysis. To fill this gap, this study proposes a hybrid approach combining latent class cluster analysis with the random parameter negative binomial regression model (LCA-RPNB) to explore the association between risk factors and bicycle crash frequency. First, the bicycle crash data is categorized into three clusters using LCA based on crash features such as gender, trip purposes, weather, and light conditions. Then, the separated crash frequency models for different clusters and the overall model are developed based on RPNB using regional factors of crash locations as independent variables and the crash frequency of different clusters respectively as dependent variables. The hybrid approach enables a comprehensive examination of internal and external heterogeneities among bicycle crash frequency factors simultaneously. Results suggest that the proposed hybrid approach exhibits superior fitting and predictive performance compared to the model only considers the effects of unobserved heterogeneity from one perspective with the lower values of Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). This approach can help policymakers and urban planners to design more effective safety interventions by understanding the distinct needs of different bicyclist clusters and the specific factors that contribute to crash risk in each group.

Modelling Passengers’ Travel Behaviour for Shared Autonomous Vehicle and Bus Considering Heterogeneity

The popularisation of autonomous vehicles will give rise to a new business model called shared autonomous vehicle (SAV). SAVs may attract a large number of passengers and lead to a decline in the share of buses, which can be interpreted by exploring passengers’ travel behaviour when confronting the SAV and bus modes. Thus, this paper addresses the SAV and bus passengers’ travel behaviour, aiming to examine the factors influencing travel behaviour and revealing the characteristics of SAV passengers. We classified passengers using latent class cluster analysis and modelled passengers’ travel behaviour based on confirmatory factor analysis and mixed logit model. The findings indicate a variation in travel preferences among different classes of travellers. Short-distance travellers pay less attention to travel time. Non-short-distance PT travellers are most likely to be affected by service attributes (waiting time, travel time and travel costs). Non-short-distance private car travellers are more likely to become early SAV adopters. Passengers travelling for short distances may be more likely to choose SAV, which reveals the potential of SAVs to become a first and last mile connection for public transport. Passengers lack trust in SAVs, which will affect their promotion.

Identifying electric vehicle charging styles among consumers: a latent class cluster analysis

The market share of electric vehicles (EVs) is growing rapidly, making it crucial to understand the charging behaviour of current and prospective users. Such understanding is essential for designing policies that positively influence consumers’ charging behaviour and facilitate EV adoption. In this study, we examined the heterogeneity in charging preferences of 994 respondents across Australia using a latent class cluster model that considers indicators of charging behaviour as outcomes of interest. We used sociodemographic characteristics, travel needs, and EV adoption status as covariates to predict class membership. Our findings indicate five segments of consumers with distinct charging preferences: routine-focused frugals, cost-oriented deliberators, range seekers, flexibility seekers, and indifferent late adopters. These segments differ in the importance they attach to charging attributes, their coping strategies with limited battery resources, and their risk attitude. Our results suggest that a uniform approach to EV-related policies is not appropriate, as each consumer segment has unique charging preferences and requirements. Furthermore, the study emphasizes the significance of accounting for charging behaviour heterogeneity in demand modelling, as assumptions in current models may not accurately represent the decision-making of most segments.

Segmenting the potential users of MaaS by combining latent class cluster analysis and structural equation modeling

To reduce problems caused by the excessive use of private vehicles, transportation planners have been actively looking for sustainable transport related solutions. Mobility as a Service (MaaS) is gaining popularity due to its capability to integrate transport opportunities into one platform. Studies on travelers’ acceptance of MaaS have been conducted, but there remains a gap in understanding heterogeneity regarding intentions of using MaaS particularly concerning attitudinal factors. Based upon the extension of the unified theory of acceptance and use of technology, we plan to reveal travelers’ segmentation by using latent class cluster analysis and to examine path heterogeneities in the identified traveler clusters by applying a structural equation modeling. The results confirm that the traveler clusters consider the importance of the relatives and media's viewpoints toward MaaS, while the car user cluster is more interested in adopting MaaS if the service could promise enough flexibility of traveling in terms of time and destination. The results related to the heterogeneity in transport mode choice are useful for constructing MaaS packages. This study could be advantageous for MaaS operators as it enables them to understand some attitudinal factors influencing willingness to take up MaaS thus realizing suitable, effective promotions.

The significance of cultural context in the classification of preferred mouth behavior: Exploring eating behavior using causal network analysis

AbstractPreferred mouth behavior studies have mostly focused on Western populations using the JBMB tool to determine how food is manipulated in the mouth. The objective of this study was to determine the relationship between preferred mouth behavior and eating behavior among the diverse group of 209 Malaysians participants. A series of Analysis of Variances and an unsupervised machine learning technique known as Latent Class Clustering were utilized to determine individual's oral behavior. The Adult Eating Behavior Questionnaire was later modeled using causal networks, specifically Partial Least Squares‐Confirmatory Factor Analysis, to uncover relationship between appetitive qualities. The results from this study showed that cultural context and appropriateness play an essential role when classifying preferred mouth behavior using behavioral questionnaires and models. Furthermore, this study also determined the relationship between eating behavior and preferred mouth behavior, as well as unique differences in eating behavior observed among the mouth behavior groups.

The algebraic learning of middle school students’ evolution during a school year: a statistical large-scale study based on results in mathematics didactics

AbstractThis large-scale study characterizes the algebraic learning of a cohort of nearly 800 French students during the last year of middle school (9th grade, 14–15 years old) and examines their evolution in relation to the composition of the classes to which the students belong. Based on the Anthropological Theory of the Didactic (ATD), the study is founded both on a reference model of elementary algebra and on technological-theoretical levels characterizing the students’ reasoning and their knowledge regarding institutional expectations. These are used to code the data and define the statistical variables on which multivariate descriptive analyses are carried out. Students’ learning is analyzed through Pépite, an algebra test given at the beginning and at the end of the year. A cluster analysis resulted in three distinct clusters of similar classes and of similar students both at the beginning and end of the grade 9 school year. They are interpreted didactically and could be useful for a teacher who has to manage the heterogeneity of learning in a class. Some classes and students move from one cluster to another between the beginning and the end of the year, showing a wide variety of ways in which students’ algebra learning progresses or regresses over a school year.

Effect of peptide-binding motif on survival of HLA-haploidentical transplantation with post-transplant cyclophosphamide.

Peptide-binding motif (PBM) model, a hierarchical clustering of HLA class I based on their binding specificity, was developed to predict immunopeptidome divergence. The effect of PBM mismatches on outcomes is unknown in HLA-haploidentical haematopoietic cell transplantation with post-transplant cyclophosphamide (PTCy-haplo). We therefore conducted a retrospective study using national registry data in PTCy-haplo. Overall, 1352 patients were included in the study. PBM-A bidirectional mismatch was associated with an increased risk of overall mortality in multivariable analysis (hazard ratio, 1.26; 95% confidence interval, 1.06 to 1.50; p = 0.010). None of relapse, non-relapse mortality (NRM) and graft-versus-host disease showed significant differences according to PBM-A bidirectional mismatch status in the entire cohort. The impact of PBM-A bidirectional mismatch on overall survival (OS) was preserved within the HLA-A genotype bidirectional mismatch population, and their lower OS stemmed from higher relapse rate in this population. The worse OS due to high NRM with PBM-A bidirectional mismatch was prominent in lymphoid malignancies receiving reduced-intensity conditioning. The PBM model may predict outcomes more accurately than HLA genotype mismatches. In conclusion, this study demonstrated that the presence of PBM-A bidirectional mismatch elevated the risk of mortality of PTCy-haplo. Avoiding PBM-A bidirectional mismatch might achieve better outcomes in PTCy-haplo.

Heterogeneity in activity and travel mode patterns of older Indonesians

Older adults aged 50 years and over may experience varied participation levels in out-of-home activities and travel modes, with their ability to walk being a significant factor. However, the heterogeneity of the activity–travel patterns of people at this age, especially in the Indonesian context, is not well understood. Using Yogyakarta as a case study and applying a latent class cluster analysis, this study is the first to categorize older adults’ participation in activities and use of travel modes and to understand the distinct characteristics of older adults. The model results revealed four discrete groups based on older adults’ activity participation and another four discrete groups based on their travel mode use. In the activity participation group, active older adults are the largest group, followed by inactive, working, and very active older adults. Meanwhile, those reliant on their motorized vehicles make up the largest travel mode group, followed by low-cost-vehicle users, captive riders of personal motorcycles, and car users. The model results also reveal that captive riders of personal motorcycles tend to be male and the youngest-old, while smartphone users are more likely to belong to the car users group. Walking ability constraints also significantly affect heterogeneity in older adults’ activity and travel mode, where those with no constraints on their ability to walk tend to be categorized as active older adult car users. The findings of this study may help policymakers identify older people with particular activity patterns and travel modes and develop policies to accommodate their mobility needs.

Developing heart stroke prediction model using deep learning with combination of fixed row initial centroid method with Navie Bayes, Decision Tree, and Artificial Neural Network

The present research and study, aimed to develop a new predictive model that easily navigate to the challenges of risk factors causing a heart stroke and accurately detect the early chances of having the stroke. Knowledge of risk factors impacting in the deterioration of a patient's health in causing the severity of heart stroke, helps future research work in prognosis of heart stoke and implement feature selection techniques by considering all such risk factors. Clustering of binary classes perfectly, without any noise, is a major advantage in predicting heart stroke patients' condition in their early stages of severity. Novel initial centroid selection method FRM is developed which considered the best until date resulting in 100 % clustering of patients into binary classes, which significantly contribute to the enhancement of accuracy results. Major objective is integrating clustering technique with classification algorithms Naïve Bayes, Decision Tree and Artificial Neural Network and built three prediction systems, FRM-NB, FRM-DT, and FRM-ANN performed the best when compared to existing systems, resulting in the best accuracy values in predicting early risk of heart stroke and reducing chances of recurrent heart strokes. We have achieved the best accuracy of 94 % (FRM-NB) and 97 % (FRM-DT), sensitivity of 90 % (FRM-NB) and 95 % (FRM-DT), specificity score of 97 % (FRM-NB) AND 90 % (FRM-DT) and AUC-ROC score of 0.976 (FRM-NB) and 0.953(FRM-DT). FRM-ANN model achieved an accuracy of 98 % with 100 % sensitivity and 0.99 score of AUC, which till date no existing research has achieved.

Exploring interacting effects of risk factors on run-off-road crash severity: An interpretable machine learning model joint with latent class clustering

Run-Off-Road (ROR) crashes are frequent and pose a significant risk of injury and fatality. Given the complexity of their mechanisms and the interaction of multiple factors, this study aims to comprehensively investigate the factors influencing the severity of ROR crashes, which have been understudied. Furthermore, addressing the current methodological challenge in machine learning (ML) crash modeling, this study proposes an approach to tackle unobserved heterogeneity in ML. An interpretable ML joint with prior latent class clustering is implemented. The significant risk factors and interactions are interpreted using SHAP (SHapley Additive exPlanations) method across clusters. This study utilizes ROR crash records, traffic, and geometric data from main suburban freeways in Iran collected over a 5-year period. The key interacting factors associated with severe ROR during adverse weather (cluster 1) are: co-occurrence of low congestion and higher speed variation; low congestion, nighttime darkness and rollover; roadway departure by buses and mini-buses and rollover occurrence; vehicle departure and collision with fixed objects. Moreover, the critical interactions for nighttime condition (cluster 2) are: curve sections combined with longitudinal slope; inside shoulder width <1.5 m and a hit median concrete NewJersey barrier. The risky interactions for crashes occurred in curves (cluster 3) are: departing in two-lane sections in low congestion conditions; vehicles collisions with the median concrete NewJersey barriers. The findings of this study enhance comprehension of the significant effects of interactions under various conditions, offering valuable insights for policymakers. Additionally, recommendations are offered to mitigate the risk of severe ROR crashes.

Analyzing shared mobility markets in Europe: A comparative analysis of shared mobility schemes across 311 European cities

The progression of shared mobility across Europe is remarkable. While station-based car and bike sharing have a more extended history, particularly in major European (capital) cities, recent advancements in modal types and operational models have significantly transformed the shared mobility landscape. Rapid expansion by private organisations has broadened access to shared mobility services across Europe. However, not all European cities are considered potentially viable markets due to local factors such as stringent regulatory frameworks and unfavourable economic conditions. The composition of the local offerings influences how citizens use these services, impacting travel behaviour and the local transport networks differently. Therefore, understanding the availability of various shared mobility schemes across Europe is essential for comprehending the market structure, its development, the providers' decision drivers, and the potential consequences for local transportation systems.First, this paper presents data on various segments and features of the shared mobility market across European cities with more than 100.000 inhabitants. Second, two cluster approaches, i.e. k-means and latent class clustering analysis (LCCA), are conducted to structure this European market. Third, the contextual characteristics, such as socio-demographics, the built environment and the geography, are compared among the clusters using Dunn testing.The results depict that the market is very fragmented, ranging from cities with a minimal offer (i.e. one type of modality available) to cities with a very competitive market consisting of numerous modalities and operators.As expected, the most comprehensive offer of shared mobility is found in cities with the highest economic potential, measured by GDP per capita and population size. However, these cities tend to impose stricter regulations and invest in public schemes, especially for bike and car sharing, affecting the share of private operators. This may explain why private scooter sharing companies are willing to operate in smaller cities that initially seem to lack the economic conditions to accommodate a profitable sharing scheme. In cities where scooters are absent, mainly in Dutch cities, free-floating moped- and bike-sharing schemes have acted as a surrogate. Still, the comprehensiveness of the offer in these cities is considerable, suggesting that even with strict regulatory frameworks, other factors like infrastructure can create an attractive environment for operators. Overall, shared mobility is well-developed in European cities, meaning that many people are already aware of or have access to some form of shared mobility. This provides opportunities for other less-developed modalities, such as cargo bicycles, to further expand and offer specific use cases for car replacement. Therefore, future research could follow up on market developments to understand how various segments evolve and to examine the role of different policy frameworks more thoroughly.