Ensemble Of Trees Research Articles

Artificial intelligence-based modeling of compressive strength of slurry infiltrated fiber concrete

Purpose The purpose of this paper is to develop a reliable model that would predict the compressive strength of slurry infiltrated fiber concrete (SIFCON) modified with various supplementary cementitious materials (SCMs) using artificial intelligence approach. Design/methodology/approach This study engaged the artificial intelligence to predict the compressive strength of SIFCON through deep neural networks (DNN), artificial neural networks, linear regression, regression trees, support vector machine, ensemble trees, Gaussian process regression and neural networks (NN). A thorough data set of 387 samples was gathered from relevant studies. Eleven variables (cement, silica fume, fly ash, metakaolin, steel slag, fine aggregates, steel fiber fraction, steel fiber aspect ratio, superplasticizer, water to binder ratio and curing ages) were taken as input to predict the output (compressive strength). The accuracy and reliability of the developed models were assessed using a variety of performance metrics. Findings The results showed that the DNN (11-20-20-20-1) predicted the compressive strength of SIFCON better than the other algorithms with R2 and mean square error yielding 95.89% and 8.07. The sensitivity analysis revealed that steel fiber, cement, silica fume, steel fiber aspect ratio and superplasticizer are the most vital variables in estimating the compressive strength of SIFCON. Steel fiber contributed the highest value to the SIFCON’s compressive strength with 16.90% impact. Originality/value This is a novel technique in predicting the compressive strength of SIFCON optimized with different SCMs using supervised learning algorithms, improving its quality and performance.

Frequency-adjusted borders ordinal forest: A novel tree ensemble method for ordinal prediction.

Ordinal responses commonly occur in psychology, e.g., through school grades or rating scales. Where traditionally parametric statistical models like the proportional odds model have been used, machine learning (ML) methods such as random forest (RF) are increasingly employed for ordinal prediction. With new developments in assessment and new data sources yielding increasing quantities of data in the psychological sciences, such ML approaches promise high predictive performance. As RF does not inherently account for ordinality, several extensions have been proposed. A promising approach lies in assigning optimized numeric scores to the ordinal response categories and using regression RF. However, these optimization procedures are computationally expensive and have been shown to yield only situational benefit. In this work, I propose Frequency-Adjusted Borders Ordinal Forest (fabOF), a novel tree ensemble method for ordinal prediction forgoing extensive optimization while offering improved predictive performance in simulation and an illustrative example of student performance. To aid interpretation, I additionally introduce a permutation variable importance measure for fabOF tailored towards ordinal prediction. When applied to the illustrative example, an interest in higher education, mother's education, and study time are identified as important predictors of student performance. The presented methodology is made available through an accompanying R package.

Evaluation of Tree-Based Voting Algorithms in Water Quality Classification Prediction

Accurately predicting the state of surface water quality is crucial for ensuring the sustainable use of water resources and environmental protection. This often requires a focus on the range of factors affecting water quality, such as physical and chemical parameters. Tree models, with their flexible tree-like structure and strong capability for partitioning and selecting influential water quality features, offer clear decision-making rules, making them suitable for this task. However, an individual decision tree model has limitations and cannot fully capture the complex relationships between all influencing parameters and water quality. Therefore, this study proposes a method combining ensemble tree models with voting algorithms to predict water quality classification. This study was conducted using five surface water monitoring sites in Qingdao, representing a portion of many municipal water environment monitoring stations in China, employing a single-factor determination method with stringent surface water standards. The soft voting algorithm achieved the highest accuracy of 99.91%, and the model addressed the imbalance in original water quality categories, reaching a Matthews Correlation Coefficient (MCC) of 99.88%. In contrast, conventional machine learning algorithms, such as logistic regression and K-nearest neighbors, achieved lower accuracies of 75.90% and 91.33%, respectively. Additionally, the model’s supervision of misclassified data demonstrated its good learning of water quality determination rules. The trained model was also transferred directly to predict water quality at 13 monitoring stations in Beijing, where it performed robustly, achieving an ensemble hard voting accuracy of 97.73% and an MCC of 96.81%. In many countries’ water environment systems, different water qualities correspond to different uses, and the magnitude of influencing parameters is directly related to water quality categories; critical parameters can even directly determine the quality category. Tree models are highly capable of handling nonlinear relationships and selecting important water quality features, allowing them to identify and exploit interactions between water quality parameters, which is especially important when multiple parameters together determine the water quality category. Therefore, there is significant motivation to develop tree model-based water quality prediction models.

Eye Movement Classification using Feature Engineering and Ensemble Machine Learning

This paper explores the classification of gaze direction using electrooculography (EOG) signals, integrating signal processing, deep learning, and ensemble learning techniques to enhance accuracy and reliability. A complex technique is proposed in which several feature types are derived from EOG data. Spectral properties generated from power spectral density analysis augment basic statistical characteristics such as mean and standard deviation, revealing the frequency content of the signal. Skewness, kurtosis, and cross-channel correlations are also used to represent intricate nonlinear dynamics and inter-channel interactions. These characteristics are then reformatted into a two-dimensional array imitating picture data, enabling the use of the pre-trained ResNet50 model to extract deep and high-level characteristics. Using these deep features, an ensemble of bagging-trained decision trees classifies gaze directions, lowering model variance and increasing prediction accuracy. The results show that the ensemble deep learning model obtained outstanding performance metrics, with accuracy and sensitivity ratings exceeding 97% and F1-score of 98%. These results not only confirm the effectiveness of the proposed approach in managing challenging EOG signal classification tasks but also imply important consequences for the improvement of Human-Computer Interaction (HCI) systems, especially in assistive technologies where accurate gaze tracking is fundamental.

RGIE: A Gene Selection Method Related to Radiotherapy Resistance in Head and Neck Squamous Cell Carcinoma.

Head and Neck Squamous Cell Carcinoma (HNSCC) is a malignant tumor with a high degree of malignancy, invasiveness, and metastasis rate. Radiotherapy, as an important adjuvant therapy for HNSCC, can reduce the postoperative recurrence rate and improve the survival rate. Identifying the genes related to HNSCC radiotherapy resistance (HNSCC-RR) is helpful in the search for potential therapeutic targets. However, identifying radiotherapy resistance-related genes from tens of thousands of genes is a challenging task. While interactions between genes are important for elucidating complex biological processes, the large number of genes makes the computation of gene interactions infeasible. We propose a gene selection algorithm, RGIE, which is based on ReliefF, Gene Network Inference with Ensemble of Trees (GENIE3) and Feature Elimination. ReliefF was used to select a feature subset that is discriminative for HNSCC-RR, GENIE3 constructed a gene regulatory network based on this subset to analyze the regulatory relationship among genes, and feature elimination was used to remove redundant and noisy features. Nine genes (SPAG1, FIGN, NUBPL, CHMP5, TCF7L2, COQ10B, BSDC1, ZFPM1, GRPEL1) were identified and used to identify HNSCC-RR, which achieved performances of 0.9730, 0.9679, 0.9767, and 0.9885 in terms of accuracy, precision, recall, and AUC, respectively. Finally, qRT-PCR validated the differential expression of the nine signature genes in cell lines (SCC9, SCC9-RR). RGIE is effective in screening genes related to HNSCC-RR. This approach may help guide clinical treatment modalities for patients and develop potential treatments.

Deep learning assisting construction of heat transfer constitutive relationships for porous media

Predicting heat transfer in porous materials is challenging due to their complex microstructure, and traditional experimental and theoretical methods often fall short. To address this, we present a machine learning-based approach using a single Descriptor-to-Property Network (D2P-Net) to identify the relationships between the effective thermal conductivity (ETC) of porous media with the structural descriptors. The D2P-Net, an ensemble of decision trees, uses eleven structural descriptors, such as porosity, pore size, and connectivity, derived from four distinct types of three-dimensional porous structures. These structures are computationally generated, and their ETC is computed using the Lattice Boltzmann Method (LBM) to create a robust dataset for training. The model is trained using mean squared error (MSE) employed as the loss function, and achieves an R2 value of 0.994 and a root mean square error (RMSE) of 0.229, indicating high predictive accuracy. Additionally, SHapley Additive exPlanations (SHAP) values are employed to analyze the importance of each structural descriptor, offering valuable insights into their contributions to ETC predictions. This approach provides a reliable and interpretable method for understanding and optimizing the thermal properties of porous materials, with potential applications in areas requiring efficient thermal management, such as hypersonic vehicles.

Incident Duration Reliability Assessment Using Monte-Carlo Simulation and Kernel Density Estimation of Machine Learning-Based Models

Traffic incidents are a major cause of non-recurrent congestion and delays, making accurate incident duration (ID) prediction essential for effective traffic management. While machine learning (ML) and deep learning (DL) models have been developed to predict ID and its sub-periods (verification, response, and clearance times), their reliability has not been systematically assessed. This study introduces a novel framework to evaluate the reliability of these predictions using 4,000 traffic incident records. Non-parametric Kernel Density Estimation effectively captured variations in the data, outperforming traditional parametric methods. Monte Carlo simulation was then used to assess prediction reliability. Among the models tested, bagged ensemble trees provided the best balance between accuracy and complexity, showing strong reliability for predicting total ID and sub-periods. Adding 5% to 25% buffer adjustments further improved reliability by accounting for prediction uncertainties. This framework offers a robust tool for assessing prediction reliability, is adaptable to various ML and DL models, and represents a significant step forward in traffic incident management.

Engression: extrapolation through the lens of distributional regression

Abstract Distributional regression aims to estimate the full conditional distribution of a target variable, given covariates. Popular methods include linear and tree ensemble based quantile regression. We propose a neural network-based distributional regression methodology called ‘engression’. An engression model is generative in the sense that we can sample from the fitted conditional distribution and is also suitable for high-dimensional outcomes. Furthermore, we find that modelling the conditional distribution on training data can constrain the fitted function outside of the training support, which offers a new perspective to the challenging extrapolation problem in nonlinear regression. In particular, for ‘preadditive noise’ models, where noise is added to the covariates before applying a nonlinear transformation, we show that engression can successfully perform extrapolation under some assumptions such as monotonicity, whereas traditional regression approaches such as least-squares or quantile regression fall short under the same assumptions. Our empirical results, from both simulated and real data, validate the effectiveness of the engression method. The software implementations of engression are available in both R and Python.

Identification of Exploited Unreliable Account Passwords in the Information Infrastructure Using Machine Learning Methods

Accounts are an integral part of most modern information systems and provide their owners with the ability to authenticate within the system. This paper presents an analysis of existing methods for detecting simple account passwords in automated systems. Their advantages and disadvantages are listed. A method was developed to detect simple exploitable passwords that administrators can use to supplement other existing methods to increase the overall security of automated systems against threats from accounts potentially compromised by attackers. The method was based on the analysis of commands executed in automated or manual modes with the indication of credentials in plain text. Minimum password strength requirements are provided based on the security level. A special case was considered in which all passwords analyzed in this way were found explicitly in the system logs. We developed a unified definition of the classification of passwords into simple and strong, and also developed machine learning technology for their classification. The method offers a flexible adaptation to a specific system, taking into account the level of significance of the information being processed and the password policy adopted, expressed in the possibility of retraining the machine learning model. The experimental method using machine learning algorithms, namely the ensemble of decision trees, for classifying passwords into strong and potentially compromised by attackers based on flexible password strength criteria, showed high results. The performance of the method is also compared against other machine learning algorithms, specifically XGBoost, Random Forest, and Naive Bayes. The presented approach also solves the problem of detecting events related to the use and storage of credentials in plain text. We used the dataset of approximately 770,000 passwords, allowing the machine learning model to accurately classify 98% of the passwords by their significance levels.

TreeHouse: An MLIR-based Compilation Flow for Real-Time Tree-based Inference

Tree-based ensembles stand as the prominent resource-efficient approaches for real-time inference. To optimize their performance, researchers have developed several solutions to accommodate their unique program structure, i.e., consecutive branches and eliminate the floating-point arithmetic operations, which are usually costly at the embedded computing systems. Most of them are realized at the level of source code and a standard compilation is applied subsequently. Therefore, an end-to-end compilation flow may consolidate these methods and provide a holistic optimization. In this work, we introduce TreeHouse, a compilation flow based on MLIR designed for real-time inference of tree ensembles. First, we optimize the layout of basic blocks to reduce the expected branches during inference. Moreover, we provide a solution to facilitate LLVM register allocation for further efficiency. In addressing the suboptimal performance of floating-point operations in edge systems, we incorporate methods to convert data into integer format. We implemented and evaluated TreeHouse using two tree-based ensembles: boosting trees and random forests. Overall, boosting trees exhibited performance gains ranging from 1.38 × to 8.24 × on x86, 1.70 × to 6.61 × on ARMv8, and 1.75 × to 4.52 × on RISC-V compared to state-of-the-art decision tree research. Random forests achieved performance gains of up to 1.6 × on x86, 2.03 × on ARMv8, and 2.9 × on RISC-V.

A hybrid machine learning-based cyber-threat mitigation in energy and flexibility scheduling of interconnected local energy networks considering a negawatt demand response portfolio

The interconnection of local energy networks (LENs) enables efficient exchange of energy and flexibility among them, fostering the integration of distributed energy resources and demand-side management strategies. Thus, the interconnected local energy systems (ILEN) structure is a viable approach to electrical distribution systems’ operation and management. However, implementing distributed energy management structures such as ILEN entails a great amount of information transactions. Therefore, these structures are more vulnerable to cyber threats. Thus, the newly developed efficient and secure power systems’ operation methods should take digitalization-related security risks into account. As a result, this paper is focused on the development of a secure operation method, equipped with a hybrid algorithm to mitigate cyber threats in the context of ILEN. In this regard, this research proposes a novel hybrid XGBoost-based cyber threat mitigation (HXGBTM) method to cope with the vulnerabilities of the physical and information layers of the cyber-infrastructure. The proposed cyber threat mitigation method is built upon the classification and regression capabilities of the XGBoost ensemble of decision trees to identify and mitigate anomalies in the electrical consumption data. Therefore, in the first step, the ILEN’s multi-objective energy and flexibility scheduling problem considering demand response portfolio i.e., MOEFSDRPILENis developed that encompasses a bi-level optimization problem, in which the operator of the ILEN optimizes energy and flexibility trading in the upper level. While in the lower level, each LEN operator minimizes scheduling costs along with maximizing the local flexibility as well as providing a demand response portfolio as a negawatt resource. Here, the flexibility index, which is later maximized using the second objective function, is considered as the proportion between "the available ramping capacity" and "required ramping capacity". In this paper, direct load control, and interruptible/curtailable demand response comprehensive models are implemented as candidate programs for the suggested portfolio. Furthermore, a hybrid cyber threat is modeled considering the communication line intrinsic vulnerability, as a result of natural causes, wear, and aging of the infrastructure etc., as well as false data injection (FDI) attacks that target each LEN’s electrical consumption database. Finally, the proposed HXGBTM is employed to mitigate the above-mentioned cyber-vulnerabilities and achieve near real-world conditions.

Abstract 4138376: A Machine Learning Approach to Predict Percutaneous Coronary Intervention in Patients with Critical Illness and Signs of Myocardial Injury

Background: Critically ill patients frequently experience multi-organ dysfunction and unstable vital signs, increasing their risk of myocardial injury and elevated serum troponin levels. Myocardial injury from oxygen supply-demand mismatch, whether due to systemic disease or obstructive coronary artery disease (CAD), often leads to patients undergoing an invasive coronary angiogram (ICA) and potentially percutaneous coronary intervention (PCI) if obstructive CAD is identified. However, this procedure carries significant risks for critically ill patients. To address this challenge, we aimed to develop a machine learning (ML) model to predict the need for PCI in critically ill patients with myocardial injury. Methods: The study cohort included critically ill patients with elevated serum troponin levels who underwent ICA during their hospitalization at a single institution from 2012 to 2022. We defined critical illness as a life-threatening condition requiring intensive support for vital sign instability. Patients with atherothrombosis identified during ICA were excluded from the study, focusing only on patients with type 2 MI. We utilized Matplotlib to analyze patient data, including demographics, lab results, vital signs, and comorbidities. We then used XGBoost, a supervised ML algorithm, for binary classification tasks. An ensemble of decision trees was employed, with gradient boosting to correct errors from previous trees. XGBoost iteratively adjusted predictions, creating a model to predict which patients required PCI. Results: Out of 765 patients, 173 (22.6%) underwent PCI. The cohort had a mean age of 60.5 ± 13.5 years, with 41.7% being female. The ML model accurately identified non-PCI candidates at 75% and PCI candidates at 64%, achieving an overall accuracy of 73%. The three most important features identified by the model were a history of diabetes, diastolic heart failure, and whether the patient received a blood transfusion during hospitalization. Conclusion: A ML approach using XGBoost demonstrated moderate accuracy in predicting the need for PCI in critically ill patients with signs of myocardial injury. This prediction holds clinical significance due to the heightened risks associated with ICA in these patients. Given that PCI occurred in only 22.6% of the cohort, future models should include a higher proportion of patients who underwent PCI to improve the robustness and true positive accuracy of the model.

Machine learning techniques for the smart faults detection and diagnosis of centrifugal compressor

In this work, we have conducted a comparative study among several machine learning techniques with the aim of selecting the best one for classifying faults affecting the compressor system to enable smart monitoring. This study encompasses various machine learning techniques, including Support Vector Machine, k-nearest neighbor, Decision Tree, Naive Bayes, AdaBoost, and Bag ensembles. To determine the optimal classification technique, we applied three distinct criteria: the confusion matrix, error histogram, and mean square error through cross-validation. Based on these criteria, the results indicate a tie for the top position between two classification models: Decision Tree and Bag ensemble. To solidify our choice of a single model, we employed the new AutoML technique to automatically identify the most suitable machine learning classification model for our case study. We evaluated this approach using process data obtained from an operational industrial centrifugal compressor. Consequently, the results presented in this work affirm that Decision Tree is the superior technique for classifying faults in the 3MCL compressor.

Predicting the probability of failure in medicinesʼ public procurement

The quality and timeliness of medicinesʼ supply to the healthcare system through public procurement is an urgent task of public policy in all countries of the world, including Russia. Failure to close (failure of tenders, termination of already concluded contracts) procurement procedures in such a socially significant area carries risks for the population, provokes the emergence of hidden transaction costs for the budget system to eliminate the consequences of procurement failures. In their previous works, the authors identified the factors that lead to the failure to close tenders for the purchase of remedies, and initially assessed the consequences of their influence on procurement procedures. The purpose of this article is to present a mathematical model of the probability of non-closure tenders based on the obtained results. To achieve this goal, through processing more than 1 million notifications of public procurement of remedies for 2022–2023, collected from the open sources, the tasks with no methodological solutions were completed. Thus, a set of features accompanying the failure to close procedures for the purchase of remedies was compiled. The composition of identified features was analyzed; their influence on non-closure of the procedure was assessed. A model of the probability of non-closure of the procedure was constructed, and the results were interpreted. Unlike previously published studies, the forecast model was implemented on the ensembles of decision trees using gradient boosting. This made possible to significantly improve the quality of the forecast for each factor affecting the probability of non-closure of the bidding. The results obtained in the article are not only scientifically novel, but can also be used by regulatory and control bodies in public procurement to develop methodological recommendations for customers on establishing optimal conditions for concluding and fulfilling contracts, which will reduce procurement risks and damage to the state budget.

Development of a recommendation engine for university study programme selection: A regression-based approach

This paper is the third in a series focused on bridging the gap between secondary and higher education. Our primary objective is to develop a robust theoretical framework for an innovative e-business model called the Undergraduate Study Programme Search System (USPSS). This system considers multiple criteria to reduce the likelihood of exam failure or the need for multiple retakes, while maximizing the chances of successful program completion. Testing of the proposed algorithm demonstrated that the Stochastic Gradient Boosted Regression Trees method outperforms the current method used in Lithuania for admitting applicants to 47 educational programs. Specifically, it is more accurate than the Probabilistic Neural Network for 25 programs, the Ensemble of Regression Trees for 24 programs, the Single Regression Tree for 18 programs, the Random Forest Regression for 16 programs, the Bayesian Additive Regression Trees for 13 programs, and the Regression by Discretization for 10 programs.

Digital twin for decision support system of industrial utility management

Manufacturing and industrial operations rely heavily on energy that is generated mostly from fossil fuels, such as coal, oil, and natural gas, which harm the environment and contribute to climate change. Thus, renewable energy is being integrated into industrial processes to lessen environmental effects and reduce fossil fuel usage. However, the renewable source performance process can be greatly affected by disturbances and constraints, such as ambient air temperature and relative humidity, minimum utility consumption, and the total energy required, making effective control difficult. This study proposes a digital twin built with machine learning regression techniques for load demand forecasting as a decision-support system for industrial utility management. From the results, the ensemble tree (ET) model produced the highest accuracy, based on validation and test dataset root mean squared error values of 23.164 and 27.558, respectively, and R2 values of 0.96 and 0.95, respectively. The digital twin and load demand forecasting approaches effectively created an efficient operating schedule for industrial utility management. The ET model had a total error of 23.86%, substantially lower than the average load demand's total error of 65.29%. Therefore, the ET model with weather conditions in four scenarios could be recommended to optimize energy utilization when creating the operating schedule.

A comprehensive evaluation of machine learning and deep learning algorithms for wind speed and power prediction

Accurate wind speed and power predictions are crucial for renewable wind energy applications. This study compares and evaluates twelve machine learning (ML) and deep learning (DL) algorithms, including single and ensemble models across various time scales from 10 min to a day and a half ahead with a particular focus on ensemble prediction algorithms. Moreover, the study proposes a wind speed and power prediction system where the outcome of the wind speed prediction (WSP) model serves as input for the wind power prediction (WPP) model. Several evaluation metrics, such as mean absolute percentage error (MAPE) and mean square error (MSE) were calculated to benchmark different model accuracies. For WSP, the extremely randomized trees, decision tree, and bagging ensemble algorithms demonstrated high accuracy across different time scales where the MAPE ranged from 3.4% to 9.2%, the MSE ranged from 0.17 to 1.15, and the adjusted coefficient of determination ranged from 94% to 99%. For WPP, bagging ensemble algorithms and extremely randomized trees were also effective for predicting different time scales where the MAPE ranged from 4.12% to 11.7% and the MSE ranged from 10945 to 2.4. Ensemble ML algorithms provide better and more accurate results than single ML algorithms. The extreme gradient boosting model showed relatively small computational time and memory according to computational cost. Moreover, this study conducted a sensitivity analysis where air pressure, wind vane, and humidity were the key predictors for WSP and WPP.

Topological embedding and directional feature importance in ensemble classifiers for multi-class classification

Cancer is the second leading cause of disease-related death worldwide, and machine learning-based identification of novel biomarkers is crucial for improving early detection and treatment of various cancers. A key challenge in applying machine learning to high-dimensional data is deriving important features in an interpretable manner to provide meaningful insights into the underlying biological mechanismsWe developed a class-based directional feature importance (CLIFI) metric for decision tree methods and demonstrated its use for the The Cancer Genome Atlas proteomics data. The CLIFI metric was incorporated into four algorithms, Random Forest (RF), LAtent VAriable Stochastic Ensemble of Trees (LAVASET), and Gradient Boosted Decision Trees (GBDTs), and a new extension incorporating the LAVA step into GBDTs (LAVABOOST). Both LAVA methods incorporate topological information from protein interactions into the decision function.The different models' performance in classifying 28 cancers resulted in F1-scores of 92.6% (RF), 92.0% (LAVASET), 89.3% (LAVABOOST) and 85.7% (GBDT), with no method outperforming all others for individual cancer type prediction. The CLIFI metric enables visualisation of the model's decision-making functions. The resulting CLIFI value distributions indicated heterogeneity in the expression of several proteins (MYH11, ERα, BCL2) across different cancer types (including brain glioma, breast, kidney, thyroid and prostate cancer) aligning with the original raw expression data.In conclusion, we have developed an integrated, directional feature importance metric for multi-class decision tree-based classification models that facilitates interpretable feature importance assessment. The CLIFI metric can be combined with incorporating topological information into the decision functions of models to introduce inductive bias, enhancing interpretability.

Predictive modeling and estimation of moisture damages in Swedish buildings: A machine learning approach

Identifying potential moisture damage is crucial for maintenance practices and assurance of well-being of occupants. However, due to limited information availability and standardization, assessing damage prevalence on the building stock scale remains understudied. By combining investigation records and building databases, this study leverages data analytic techniques and machine learning modeling to characterize damage pathology and predict its occurrence in Swedish buildings. The interrelationships between damage-specific attributes and their associations with building parameters of several damage types were analyzed using feature selection, forming the basis for developing predictive models. Results show that multilabel classifiers outperform binary classifiers for every damage type, with lead tree ensemble models achieving minimum average AUCPR and F2 of 0.85 for microbial growth, 0.87 for deformation, 0.91 for odor, and 0.95 for water leakage. The identified patterns were interpreted and verified against descriptive statistics. The binary relevance models estimate that one-third of school buildings, 20% of commercial and office buildings, and 15% of residential dwellings in regional building stock contain moisture damage. These findings advance the quantification of moisture damage by providing new knowledge and approaches for appraising moisture damage likelihood at aggregated and individual building levels, thereby aiding in moisture safety evaluations and preventive maintenance efforts.

Inherently interpretable machine learning for credit scoring: Optimal classification tree with hyperplane splits

An accurate and interpretable credit scoring model plays a crucial role in helping financial institutions reduce losses by promptly detecting, containing, and preventing defaulters. However, existing models often face a trade-off between interpretability and predictive accuracy. Traditional models like Logistic Regression (LR) offer high interpretability but may have limited predictive performance, while more complex models may improve accuracy at the expense of interpretability. In this paper, we tackle the credit scoring problem with imbalanced data by proposing two new classification models based on the optimal classification tree with hyperplane splits (OCT-H). OCT-H provides transparency and easy interpretation with ‘if-then’ decision tree rules. The first model, the cost-sensitive optimal classification tree with hyperplane splits (CSOCT-H). The second model, the optimal classification tree with hyperplane splits based on maximizing F1-Score (OCT-H-F1), aims to directly maximize the F1-score. To enhance model scalability, we introduce a data sample reduction method using data binning and feature selection. We then propose two solution methods: a heuristic approach and a method utilizing warm-start techniques to accelerate the solving process. We evaluated the proposed models on four public datasets. The results show that OCT-H significantly outperforms traditional interpretable models, such as Decision Trees (DT) and Logistic Regression (LR), in both predictive performance and interpretability. On certain datasets, OCT-H performs as well as or better than advanced ensemble tree models, effectively narrowing the gap between interpretable models and black-box models.