Tree-based Ensemble Models Research Articles

Enhancing bond strength prediction at UHPC-NC interface: A data-driven approach with augmentation and explainability

Existing concrete structures often have difficulty reaching their design service life due to aging, increased loads, and natural disasters, and they need to be repaired and strengthened or replaced. Ultra-high-performance concrete (UHPC), known for its ultra-high strength and excellent toughness, offers a promising solution for repairing and strengthening normal concrete (NC) structures. It is expected to be a viable solution when applied to the repair and strengthening of NC structures. A reliable interfacial bonding between the two materials (NC substrate and UHPC) determines the overall performance of this composite structure. In this study, a data-driven approach is proposed to predict the bond performance at the UHPC-NC interface as accurately as possible. To overcome the problem of limited experimental data, a data augmentation model is introduced, combining kernel density estimation (KDE) and tabular generative adversarial networks (TGAN). The optimal model is determined from six decision tree-based ensemble learning models applied to two strategies: "synthetic training - real testing" and "real training - real testing." Finally, a parametric analysis is performed using SHapley Additive exPlanations (SHAP) to elucidate the importance and sensitivity of different features related to bond strength. The findings illustrate that the suggested KDE-TGAN model effectively captures the distribution of the original dataset, enhancing both the accuracy and robustness of the bond strength prediction models. Furthermore, the model's ability to explain the importance and sensitivity of different features provides valuable insights into bond strength prediction. Thus, the proposed data augmentation model provides a reliable approach for modeling experimental data with small samples in structural engineering applications.

Early prediction of mortality and morbidities in VLBW preterm neonates using machine learning.

Predicting mortality and specific morbidities before they occur may allow for interventions that may improve health trajectories. Integrating key maternal and postnatal infant variables in the first 2 weeks of age into machine learning (ML) algorithms will reliably predict survival and specific morbidities in VLBW preterm infants. ML algorithms were developed to integrate 47 features for predicting mortality, bronchopulmonary dysplasia (BPD), neonatal sepsis, necrotizing enterocolitis (NEC), intraventricular hemorrhage (IVH), cystic periventricular leukomalacia (PVL), and retinopathy of prematurity (ROP). A retrospective cohort (n = 3341) was used to train and validate the models with a repeated 10-fold cross-validation strategy. These models were then tested on a separate cohort (n = 447) to evaluate the final model performance. Among the seven ML algorithms employed, tree-based ensemble models, specifically Random Forest (RF) and XGBoost, had the best performance metrics. The area under the receiver operating characteristic curve (AUROC) of sepsis with or without meningitis (0.73), NEC (0.73), BPD (0.71), and mortality (0.74) exceeded 0.7, while the area under Precision-Recall curve (AUPRC) for all outcomes was greater than the prevalence, demonstrating effective risk stratification in VLBW preterm infants. Our study demonstrates the potential of predictive analytics leveraging ML techniques in advancing precision medicine. Reliable prediction of adverse outcomes before they occur has the potential to institute interventions and possibly improve health trajectories in VLBW preterm infants. We used machine learning to develop and test predictive models for mortality and five major morbidities in VLBW preterm infants. Individualized prediction of outcomes and individualized interventions will advance Precision Medicine in Neonatology.

Discovering the underground coal mining accident patterns in Spain from 2003 to 2021: Insights through machine learning techniques

The safety of underground coal mining has always been a global concern, involving the stable supply of energy and stakes in miners’ lives. Lessons learned from historical accidents and transforming into practical experience help reduce the quantity and severity of accidents. In this study, six ensemble learning techniques, including AdaBoost, Extra Trees, GBDT, LightGBM, Random Forest, and XGBoost, were used to investigate the correlation between accident-causing factors and severity. Firstly, 39487 underground coal mine accidents data was obtained from Spain, variables were categorized and coded. To address the extreme class imbalance, a new dataset (2468 cases) was obtained by data sampling from the original database. Subsequently, the new dataset was randomly divided into training sets (75% of the data) and test sets (25% of the data), then the hyperparameters of each model were optimized and configured. Thirdly, the models’ performance was evaluated on the test data by five metrics (accuracy, Cohen’s Kappa, precision, recall, and F1). Finally, accident patterns were derived from the identified variables along with preventive strategies. Results show that tree-based ensemble learning model performs better compared to the boosting model, and the relative importance of seven variables were determined, where previous cause (PC) and material agent (MA) are the most important factors, followed by the miner’s physical activity (PA), age (A), and experience (E), scale (S) and preventive organization (PO) are in the third tier. Furthermore, the type of accident and injury caused by PC were confirmed. Working with hand tools, younger age, lack of experience, small-scale coal mines, and unfit preventive organization increased the risk of accidents. This study not only facilitates the prediction of accident severity but also provides strategies for preventing and mitigating accidents.

WebMRT: An online tool to predict summertime mean radiant temperature using machine learning

Mean Radiant Temperature (Tmrt) is the most critical atmospheric variable influencing outdoor human thermal exposure and comfort in hot, dry environments. However, accurately quantifying Tmrt requires time-consuming field measurements with expensive equipment or complex, resource-intensive computations. We introduce WebMRT, an online tool to predict Tmrt using a data-driven approach. It features an intuitive interface using air temperature, shading status, and built environment features as predictors of Tmrt for a user-selected summer day, time, and location. Utilizing a tree-based ensemble model, WebMRT is trained on state-of-the-art human-biometeorological data collected by MaRTy using LightGBM after evaluating its performance against several candidate machine learning regressors. Feature engineering was applied to the day and time input, and two additional temporal features were derived: ‘Solar Altitude’ and ‘Minutes-from-Sunrise’. These inputs are integrated into the user interface, emphasizing simplicity and easy access for users at the frontend. After training the regressor on MaRTy datasets and employing k-fold cross-validation with ten folds, the model demonstrated strong predictive power (R2=0.92) with acceptable error (RMSE=3.43, MAPE=5.33) and bias (MBE=0.20). WebMRT also features optional fisheye photo uploads, processed using transfer learning techniques for image segmentation, further enhancing the tool's predictive accuracy, user experience, and applications towards climate action decision-making processes.

Explainable AI based slip prediction of steel-UHPC interface connected by shear studs

Recently, ultrahigh performance concrete (UHPC)-steel composite structures with shear studs have garnered interest for their potential high ductility and resilience in infrastructure. Ensuring their ductility is crucial for failure prediction and resilient infrastructure development. However, pull-out tests show that studs in UHPC often fail to meet a minimum relative slip of 6 mm, with no established direct methods for predicting this slip. This study proposes a novel approach to evaluate the shear slip at the UHPC-steel interface connected by shear studs, utilizing explainable artificial intelligence. We integrate a Light Gradient Boosting Machine (LightGBM) algorithm with SHapley Additive exPlanations (SHAP) to predict and interpret shear slip. The model development utilizes a dataset of 184 instances from UHPC-steel pullout tests, encompassing 16 features related to UHPC slab characteristics, shear pockets, and stud shear connectors. Engineered features and shear strength are strategically incorporated as predictors under four scenarios to achieve the most optimal model. The best performing model achieves an RMSE of 0.814 mm, an MAE of 0.598 mm, and an R2 of 0.838. The model’s superiority is reaffirmed through comparative analysis against five alternative tree-based ensemble models. SHAP analysis demonstrates that features related to the UHPC slab and stud shear connectors contribute 65 % and 35 % respectively to slip prediction. The top influential features are dst, sl and fc from the stud, and fu from the slab. The engineered features and ultimate load per stud indicate limited efficacy in improving model precision. Higher values of dst and fu positively contribute to shear slip prediction, whereas higher values of sl negatively impact shear slip prediction. Based on several SHAP analyses, a set of parameters for the potential ductile design of stud shear connectors in UHPC slabs is proposed. Overall, the integration of LightGBM and SHAP maintains high generalizability and enhances model transparency, instilling confidence for practical deployment and promising tangible advancements in real-world structural applications.

Enhancing Flight Delay Predictions Using Network Centrality Measures

Accurately predicting flight delays remains a significant challenge in the aviation industry due to the complexity and interconnectivity of its operations. The traditional prediction methods often rely on meteorological conditions, such as temperature, humidity, and dew point, as well as flight-specific data like departure and arrival times. However, these predictors frequently fail to capture the nuanced dynamics that lead to delays. This paper introduces network centrality measures as novel predictors to enhance the binary classification of flight arrival delays. Additionally, it emphasizes the use of tree-based ensemble models, specifically random forest, gradient boosting, and CatBoost, which are recognized for their superior ability to model complex relationships compared to single classifiers. Empirical testing shows that incorporating centrality measures improves the models’ average performance, with random forest being the most effective, achieving an accuracy rate of 86.2%, surpassing the baseline by 1.7%.

Forecasting regional in-situ thermal conductivity of soil based on tree-based ensemble learning

Borehole heat exchanger is essential component in ground-source heat pumps (GSHPs) for developing geothermal energy, where device efficiency depends on the thermal conductivity of soil. Although the thermal response test is generally used to measure this parameter, its time-consuming and susceptibility to test conditions can impact the accuracy of the test. To address this issue, the study conducted shallow geothermal surveys to establish a comprehensive dataset, enabling the construction of the in-situ thermal conductivity predictive model. By selecting seven input features, four classic tree-based ensemble learning models and stacking algorithms were used. Among all the post-tuned models examined, XGBoost-RF emerged as the standout performer, boasting an impressive R2 value of 0.9809 along with the lowest RMSE and MAE. Notably, it demonstrated generalization capabilities, with errors consistently staying within 5 % during the validation process. Additionally, the SHapley Additive exPlanations (SHAP) was used to enhance the interpretability of the model. Among these, the weighted thermal conductivity of the geotechnical was the most influential feature, while groundwater conditions such as permeability coefficient and aquifer thickness were also significant factors. This study provides a rapid and precise prediction of regional in-situ thermal conductivity. It leads to effective decision-making in the design stage of GSHPs.

Explainable AI and tree-based ensemble models: a comparative study in predicting chemical pulmonary toxicity

Explainable AI and tree-based ensemble models: a comparative study in predicting chemical pulmonary toxicity

Predicting Cd accumulation in crops and identifying nonlinear effects of multiple environmental factors based on machine learning models

The traditional prediction of the Cd content in grains (Cdg) of crops primarily relies on the multiple linear regression models based on soil Cd content (Cds) and pH, neglecting inter-factorial interactions and nonlinear causal links between external environmental factors and Cdg. In this study, a comprehensive index system of multi-type environmental factors including soil properties, geology, climate, and anthropogenic activity was constructed. The machine learning models of the tree-based ensemble, support vector regression, artificial neural network for predicting Cdg of rice and wheat based on the environmental factor indexes significantly improved the accuracy than the traditional models of linear regression based on soil properties. Among them, the tree-based ensemble models of XGboost and random forest exhibited highest accuracies for predicting Cdg of rice and wheat, with R2 in the test dataset of 0.349 and 0.546, respectively. This study found that soil properties, including Cds, pH, and clay, have greater impacts on Cdg of rice and wheat, with combined contribution rates accounting for 65.2 % and 29.7 % respectively. Since wheat sampling areas are located in central and northern China, they are more constrained by precipitation and temperature than rice sampling areas in the south. Geologic and climate factors have a greater impact on Cdg of wheat, with a combined contribution rate of 49.9 %, which is higher than the corresponding rate of 20.9 % in rice. Furthermore, the Cdg of rice and wheat did not exhibit an absolute linear relationship with Cds, and excessively high Cds can reduce the bioconcentration factor of Cd accumulation in crops. Meanwhile, other environmental factors such as temperature, precipitation, elevation have marginal effects on the increase of Cdg of crops. This study provides a novel framework to optimize traditional soil plant transfer models, as well as offer a step towards realizing high precision prediction of Cd content in crops.

Addressing bias in bagging and boosting regression models.

As artificial intelligence (AI) becomes widespread, there is increasing attention on investigating bias in machine learning (ML) models. Previous research concentrated on classification problems, with little emphasis on regression models. This paper presents an easy-to-apply and effective methodology for mitigating bias in bagging and boosting regression models, that is also applicable to any model trained through minimizing a differentiable loss function. Our methodology measures bias rigorously and extends the ML model's loss function with a regularization term to penalize high correlations between model errors and protected attributes. We applied our approach to three popular tree-based ensemble models: a random forest model (RF), a gradient-boosted model (GBT), and an extreme gradient boosting model (XGBoost). We implemented our methodology on a case study for predicting road-level traffic volume, where RF, GBT, and XGBoost models were shown to have high accuracy. Despite high accuracy, the ML models were shown to perform poorly on roads in minority-populated areas. Our bias mitigation approach reduced minority-related bias by over 50%.

SleepBoost: a multi-level tree-based ensemble model for automatic sleep stage classification.

Neurodegenerative diseases often exhibit a strong link with sleep disruption, highlighting the importance of effective sleep stage monitoring. In this light, automatic sleep stage classification (ASSC) plays a pivotal role, now more streamlined than ever due to the advancements in deep learning (DL). However, the opaque nature of DL models can be a barrier in their clinical adoption, due to trust concerns among medical practitioners. To bridge this gap, we introduce SleepBoost, a transparent multi-level tree-based ensemble model specifically designed for ASSC. Our approach includes a crafted feature engineering block (FEB) that extracts 41 time and frequency domain features, out of which 23 are selected based on their high mutual information score (> 0.23). Uniquely, SleepBoost integrates three fundamental linear models into a cohesive multi-level tree structure, further enhanced by a novel reward-based adaptive weight allocation mechanism. Tested on the Sleep-EDF-20 dataset, SleepBoost demonstrates superior performance with an accuracy of 86.3%, F1-score of 80.9%, and Cohen kappa score of 0.807, outperforming leading DL models in ASSC. An ablation study underscores the critical role of our selective feature extraction in enhancing model accuracy and interpretability, crucial for clinical settings. This innovative approach not only offers a more transparent alternative to traditional DL models but also extends potential implications for monitoring and understanding sleep patterns in the context of neurodegenerative disorders. The open-source availability of SleepBoost's implementation at https://github.com/akibzaman/SleepBoost can further facilitate its accessibility and potential for widespread clinical adoption.

Mesosilicate materials for environmental applications: Adsorption separation analysis via hybrid computational and machine learning tools

In this extensive analysis, we delve into the utilization of advanced hybrid model for molecular separation description via mesoporous materials. The process considered here is adsorption which is utilized for molecular separation in liquid solution. Data for adsorption are obtained in terms of solute concentration by solving mass transfer equations numerically. Then, the data is used for training and validation of several machine learning techniques to establish the hybrid model. For machine learning model, several methods are utilized including three robust tree-based ensemble models, namely Gradient Boosting (GB), Random Forest (RF), and Extra Trees (ET). The methos were utilized to tackle the challenging task of predicting chemical concentrations in a dataset comprising an array of over 19,000 data. These data points are characterized by input variables representing coordinates (x and y), while the output variable encapsulates chemical concentration (C). By subjecting these models to a meticulous optimization process, we achieved substantial enhancements in their predictive capabilities. RF emerged as the standout performer, boasting a remarkable R2 score of 0.99902. ET and GB also demonstrated commendable performance improvements, underscoring the versatility of tree-based ensemble models in delivering precise and reliable predictions for chemical concentrations within intricate and multifaceted datasets.

Machine Learning Modeling Integrating Experimental Analysis for Predicting Compressive Strength of Concrete Containing Different Industrial Byproducts

This study aimed to develop accurate models for estimating the compressive strength (CS) of concrete using a combination of experimental testing and different machine learning (ML) approaches: baseline regression models, boosting model, bagging model, tree-based ensemble models, and average voting regression (VR). The research utilized an extensive experimental dataset with 14 input variables, including cement, limestone powder, fly ash, granulated glass blast furnace slag, silica fume, rice husk ash, marble powder, brick powder, coarse aggregate, fine aggregate, recycled coarse aggregate, water, superplasticizer, and voids in mineral aggregate. To evaluate the performance of each ML model, five metrics were used: mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), coefficient of determination (R2-score), and relative root mean squared error (RRMSE). The comparative analysis revealed that the VR model exhibited the highest effectiveness, displaying a strong correlation between actual and estimated outcomes. The boosting, bagging, and VR models achieved impressive R2-scores in the range of 86.69%–92.43%, with MAE ranging from 3.87 to 4.87, MSE from 21.74 to 38.37, RMSE from 4.66 to 4.87, and RRMSE between 8% and 11%. Particularly, the VR model outperformed all other models with the highest R2-score (92.43%) and the lowest error rate. The developed models demonstrated excellent generalization and prediction capabilities, providing valuable tools for practitioners, researchers, and designers to efficiently evaluate the CS of concrete. By mitigating environmental vulnerabilities and associated impacts, this research can significantly contribute to enhancing the quality and sustainability of concrete construction practices.

Measuring and classifying IP usage scenarios: a continuous neural trees approach

Understanding user behavior via IP addresses is a crucial measure towards numerous pragmatic IP-based applications, including online content delivery, fraud prevention, marketing intelligence, and others. While profiling IP addresses through methods like IP geolocation and anomaly detection has been thoroughly studied, the function of an IP address—e.g., whether it pertains to a private enterprise network or a home broadband—remains underexplored. In this work, we initiate the first attempt to address the IP usage scenario classification problem. We collect data consisting of IP addresses from four large-scale regions. A novel continuous neural tree-based ensemble model is proposed to learn IP assignment rules and complex feature interactions. We conduct extensive experiments to evaluate our model in terms of classification accuracy and generalizability. Our results demonstrate that the proposed model is capable of efficiently uncovering significant higher-order feature interactions that enhance IP usage scenario classification, while also possessing the ability to generalize from the source region to the target one.

Sepsis Prediction at Emergency Department Triage Using Natural Language Processing: Retrospective Cohort Study.

Despite its high lethality, sepsis can be difficult to detect on initial presentation to the emergency department (ED). Machine learning-based tools may provide avenues for earlier detection and lifesaving intervention. The study aimed to predict sepsis at the time of ED triage using natural language processing of nursing triage notes and available clinical data. We constructed a retrospective cohort of all 1,234,434 consecutive ED encounters in 2015-2021 from 4 separate clinically heterogeneous academically affiliated EDs. After exclusion criteria were applied, the final cohort included 1,059,386 adult ED encounters. The primary outcome criteria for sepsis were presumed severe infection and acute organ dysfunction. After vectorization and dimensional reduction of triage notes and clinical data available at triage, a decision tree-based ensemble (time-of-triage) model was trained to predict sepsis using the training subset (n=950,921). A separate (comprehensive) model was trained using these data and laboratory data, as it became available at 1-hour intervals, after triage. Model performances were evaluated using the test (n=108,465) subset. Sepsis occurred in 35,318 encounters (incidence 3.45%). For sepsis prediction at the time of patient triage, using the primary definition, the area under the receiver operating characteristic curve (AUC) and macro F1-score for sepsis were 0.94 and 0.61, respectively. Sensitivity, specificity, and false positive rate were 0.87, 0.85, and 0.15, respectively. The time-of-triage model accurately predicted sepsis in 76% (1635/2150) of sepsis cases where sepsis screening was not initiated at triage and 97.5% (1630/1671) of cases where sepsis screening was initiated at triage. Positive and negative predictive values were 0.18 and 0.99, respectively. For sepsis prediction using laboratory data available each hour after ED arrival, the AUC peaked to 0.97 at 12 hours. Similar results were obtained when stratifying by hospital and when Centers for Disease Control and Prevention hospital toolkit for adult sepsis surveillance criteria were used to define sepsis. Among septic cases, sepsis was predicted in 36.1% (1375/3814), 49.9% (1902/3814), and 68.3% (2604/3814) of encounters, respectively, at 3, 2, and 1 hours prior to the first intravenous antibiotic order or where antibiotics where not ordered within the first 12 hours. Sepsis can accurately be predicted at ED presentation using nursing triage notes and clinical information available at the time of triage. This indicates that machine learning can facilitate timely and reliable alerting for intervention. Free-text data can improve the performance of predictive modeling at the time of triage and throughout the ED course.

Computational modeling of nanofluid heat transfer using Fuzzy-based bee algorithm and machine learning method

A hybrid computational method via CFD (computational fluid dynamics) and artificial intelligence was developed for studying heat transfer in nanofluid. The finite volume approach was used to solve the fluid flow equations through a pipe containing CuO nanofluid and the obtained temperature distribution was used to develop the artificial intelligence-based models. Optimization of tree-based ensemble models for predicting temperature values based on the CFD input features of x-coordinate and y-coordinate were carried out. The tree-based ensembles, i.e., Random Forest, Gradient Boosting, Extra Tree, and Adaboost Decision Trees, are optimized using the Fuzzy-based Bee Algorithm (FBA) to enhance their performance. The Extra Tree model achieved an excellent fitting with an R2 criterion of 0.99779, and RMSE of 1.7977E-01. The Random Forest model achieved slightly higher correlative accuracy, with an R2 criterion of 0.99865, and RMSE value of 1.4336E-01. The Adaboost Decision Trees model also demonstrated strong performance, with an R2 criterion of 0.99846, and RMSE of 1.5394E-01. However, the Gradient Boosting model exhibited slightly lower fitting accuracy, with an R2 score of 0.99258, and RMSE of 3.4090E-01.

Prediction of robo-advisory acceptance in banking services using tree-based algorithms.

The banking sector is increasingly recognising the need to implement robo-advisory. The introduction of this service may lead to increased efficiency of banks, improved quality of customer service, and a strengthened image of banks as innovative institutions. Robo-advisory uses data relating to customers, their behaviors and preferences obtained by banks from various communication channels. In the research carried out in the work, an attempt was made to obtain an answer to the question whether the data collected by banks can also be used to determine the degree of consumer interest in this type of service. This is important because the identification of customers interested in the service will allow banks to direct a properly prepared message to a selected group of addressees, increasing the effectiveness of their promotional activities. The aim of the article is to construct and examine the effectiveness of predictive models of consumer acceptance of robo-advisory services provided by banks. Based on the authors' survey on the use of artificial intelligence technology in the banking sector in Poland, in this article we construct tree-based models to predict customers' attitudes towards using robo-advisory in banking services using, as predictors, their socio-demographic characteristics, behaviours and attitudes towards modern digital technologies, experience in using banking services, as well as trust towards banks. In our study, we use selected machine learning algorithms, including a decision tree and several tree-based ensemble models. We showed that constructed models allow to effectively predict consumer acceptance of robo-advisory services.

Predicting the Hardness of Al-Sc-X Alloys with Machine Learning Models, Explainable Artificial Intelligence Analysis and Inverse Design

In this study, the Vickers hardness of precipitation-strengthened Al-Sc-X (X = Zr, Si, and Fe) alloys were predicted using machine learning models, depending on the alloys’ compositions, solid-solution treatment and aging conditions. The data used for machine learning were collected from the literature. Among the models, tree-based ensemble models such as extreme gradient boosting and random forest performed well. Then the feature impact on the model output was analyzed with SHarpely Additive eXplanation (SHAP). Based on the SHAP analysis and prior domain knowledge, the process conditions were restricted to narrow down the inverse design search space. Candidate alloys suggested by the optimization using a genetic algorithm showed improved hardness values. The hardness prediction model and the inverse designsuggested candidates were then experimentally validated. The accuracy of the hardness prediction model was 0.994, when the predicted hardness was 85.4 Hv, and the experimentally measured hardness was 84.9 Hv. A specimen whose composition was close to the inverse-designed alloy was cast and heat treated according to the suggested conditions. The inverse design showed an accuracy of 0.965. Exploring the entire combination of possible feature space requires vast effort and time. An efficient search for materials with improved properties can be achieved using an appropriate configuration of well-performing machine learning models and explainable AI techniques guided by domain knowledge.

Tree-Based Ensemble Models, Algorithms and Performance Measures for Classification

Tree-Based Ensemble Models, Algorithms and Performance Measures for Classification

Performance of Decision-Tree-Based Ensemble Classifiers in Predicting Fog Frequency in Ungauged Areas

Abstract Fog is a phenomenon that exerts significant impacts on transportation, aviation, air quality, agriculture, and even water resources. While data-driven machine learning algorithms have shown promising performance in capturing nonlinear fog events at point locations, their applicability to different areas and time periods is questionable. This study addresses this issue by examining five decision-tree-based classifiers in a South Korean region, where diverse fog formation mechanisms are at play. The five machine learning algorithms were trained at point locations and tested with other point locations for time periods independent of the training processes. Using the ensemble classifiers and high-resolution atmospheric reanalysis data, we also attempted to establish fog occurrence maps in a regional area. Results showed that machine learning models trained on the local datasets exhibited superior performance in mountainous areas, where radiative cooling predominantly contributes to fog formation, compared to inland and coastal regions. As the fog generation mechanisms diversified, the tree-based ensemble models appeared to encounter challenges in delineating their decision boundaries. When they were trained with the reanalysis data, their predictive skills were significantly decreased, resulting in high false alarm rates. This prompted the need for postprocessing techniques to rectify overestimated fog frequency. While postprocessing may ameliorate overestimation, caution is needed to interpret the resultant fog frequency estimates, especially in regions with more diverse fog generation mechanisms. The spatial upscaling of machine learning–based fog prediction models poses challenges owing to the intricate interplay of various fog formation mechanisms, data imbalances, and potential inaccuracies in reanalysis data.