Interpretable Tree‐Based Models for Predicting Short‐Term Rockburst Risk Considering Multiple Factors

Rockburst is a serious hazard in underground engineering, and accurate prediction of rockburst risk is challenging. To construct an intelligent prediction model of rockburst risk with interpretability and high accuracy, three binary scorecards predicting different risk levels of rockburst were constructed using ChiMerge, evidence weight theory, and the logistic regression algorithm. An intelligent rockburst prediction model based on scorecard methodology (IRPSC) was obtained by integrating the three scorecards. The effects of hazard sample category weights on the missed alarm rate, false alarm rate, and accuracy of the IRPSC were analyzed. Results show that the accuracy, false alarm rate, and missed alarm rate of the IRPSC for rockburst prediction in riverside hydropower stations are 75%, 12.5%, and 12.5%, respectively. Setting higher hazard sample category weights can reduce the missed alarm rate of IRPSC, but it will lead to a higher false alarm rate. The IRPSC can adaptively adjust the threshold and weight value of the indicator and convert the abstract machine learning model into a tabular form, which overcomes the commonly black box problems of machine learning model, as well as is of great significance to the application of machine learning in rockburst risk prediction.

Since the first recorded rock burst occurred in England in 1738, more than 20 countries have reported rock bursts (Kabiesz and Makowka 2009; Ortlepp and Stacey 1994; Patynska and Kabiesz 2009; Patynska 2013; Uszko 2009), including Germany, South Africa, Poland, the Czech Republic, Canada, Japan, France, etc. In China, rock bursts have become a common safety issue in underground coal mining. The number of coal mines experiencing rock bursts has increased annually (Jiang et al. 2010; Li et al. 2015). To date, 142 coal mines in China have suffered rock bursts which resulted in large economic losses and heavy casualties. For instance, the rock burst on 15 March 2013 in Junde Coal mine, Hegang City, caused the closure of a 200 m gateway, trapped 24 people, and killed four (Lu et al. 2015). The rock burst on 3 November 2011 in Qianqiu Coal mine, Yima City trapped 75 people underground and killed 10 people (Li et al. 2015). The rock burst on 14 February 2005 in Sunjiawan Coal mine, Fuxin City, caused a serious gas explosion and killed 214 people (State Administration of Work Safety, State Administration of Coalmine Safety 2005). The temporal and spatial evolution of mining-induced tremors reveals the process of initiation, development, and expansion of micro-fractures inside the coal-rock mass together with energy accumulation and release. This process may develop to cause either a tremor or a rock burst (Li et al. 2014; Wang et al. 2013). Tremors, rock bursts, or both are more likely to be induced in high-stress regions and the energy released therein is much higher. Therefore, rock burst risk can be evaluated by locating high-stress, and high-energy, regions. Microseismic (MS) monitoring is applicable when detecting the location and energy of mining-induced tremors. Recently, seismic velocity tomography (SVT) has been widely used for inference of high-stress distribution zones in underground mines by introducing seismic signals received by MS monitoring systems. For instance, Luxbacher et al. (2008) and Hosseini et al. (2012, 2013) conducted SVT by introducing mining-induced seismic signals, and found that high-velocity regions agreed well with high-abutment stress regions as predicted by numerical modelling, both of which were observed to redistribute as the coalface advanced. Dou et al. (2012), Banka and Jaworski (2010), and Lurka (2008) conducted SVT at regular time intervals during longwall mining, and found that rock bursts, or strong seismic events (i.e. tremors in underground mining), mainly occurred in high-velocity regions. Meanwhile, the bursting strain energy (BSE) index, which views rock bursts as a process of energy accumulation and release in the coal-rock mass as mining activities disturb the in situ stress field, was proposed here to characterise the spatial distribution of tremors [refer to Cai et al. (2015) for more details]. It was found that the & Si-yuan Gong libcumt@163.com

Recently, the spread of white spot disease in shrimps has a major impact on the aquaculture activity worldwide affecting the economy of the countries, especially South-East Asian countries like Vietnam. This deadly disease in shrimps is caused by the White Spot Syndrome Virus (WSSV). Researchers are trying to understand the spread and control of this disease by doing field and laboratory studies considering effect of environmental conditions on shrimps affected by WSSV. Generally, they have not considered spatial factors in their study. Therefore, in the present study, we have used spatial (distances to roads and factories) as well as physio-chemical factors of water: Chemical Oxygen Demand (COD), Dissolved Oxygen (DO), Salinity, NO3, P3O4 and pH, for developing WSSV susceptibility maps of the area using Decision Tree (DT)-based Machine Learning (ML) models namely Random Tree (RT), Extra Tree (ET), and J48. Model’s performance was evaluated using standard statistical measures including Area Under the Curve (AUC). The results indicated that ET model has the highest accuracy (AUC: 0.713) in predicting disease susceptibility in comparison to other two models (RT: 0.701 and J48: 0.641). The WSSV susceptibility maps developed by the ML technique, using DT (ET) method, will help decision makers in better planning and control of spatial spread of WSSV disease in shrimps.

This study presents a comparative framework for evaluating the predictive performance of four machine learning models namely Gradient Boosting Machine (GBM), Random Decision Forest (RDF), Non-Parametric Regression (NP), and Decision Tree (TREE), in estimating the Unconfined Compressive Strength (UCS) of nano-doped fly ash reinforced clayey soil. The key innovation lies in combining ensemble learning with sensitivity, monotonicity, and SHAP (SHapley Additive exPlanations) analyses to enhance predictive accuracy and interpretability in geotechnical applications. Using a comprehensive dataset of key variables (Curing Days, Maximum Dry Density (MDD), Optimum Moisture Content (OMC), Fly Ash, Multi-Walled Carbon Nanotubes (MWCNT), and Sodium Hexametaphosphate (SHMP)) models were trained and validated using various statistical metrics (R², MAE, MSE etc.). GBM achieved the best performance (R²: 1.000, 0.955; MAE: 0.001, 0.022; MSE: 0.000, 0.001 in training and testing respectively), consistently outperforming RDF, NP, and TREE. Taylor diagrams, REC curves, and AOC analysis further confirmed GBM's superior generalization and minimal error rates. Sensitivity analysis identified Days (0.4556), MDD (0.2458), and SHMP (0.1558) as the most influential factors, trends that were corroborated by SHAP analysis. Monotonicity analysis validated positive relationships with Days (R² = 0.9845) and MDD (R² = 0.689), while OMC and SHMP showed inverse effects. These results establish GBM as a highly accurate and interpretable model for UCS prediction, offering a powerful tool for optimizing soil stabilization in construction and geotechnical engineering.

Simple SummaryEffective models for predicting non-functioning pituitary neuro-endocrine tumours (NF PitNET) recurrence and regrowth following surgical intervention remain elusive. Previous studies have identified conflicting risk factors in predictions of tumour progression for patients receiving surgery for NF PitNET. The aim of this study was to develop machine learning (ML) models to improve prediction of post-operative NF PitNET progression up to 15 years following surgery. ML models were shown to be effective for predicting tumour remission, stability, and regrowth, but were non-performant when predicting tumour recurrence or reduction in size. The extent of surgical resection was shown to have the strongest influence in the performant models, with lesser influence from age, tumour volume, and the use of post-operative radiotherapy, and with no influence shown from pre- or post-operative endocrine function.Post-operative tumour progression in patients with non-functioning pituitary neuroendocrine tumours is variable. The aim of this study was to use machine learning (ML) models to improve the prediction of post-operative outcomes in patients with NF PitNET. We studied data from 383 patients who underwent surgery with or without radiotherapy, with a follow-up period between 6 months and 15 years. ML models, including k-nearest neighbour (KNN), support vector machine (SVM), and decision tree, showed superior performance in predicting tumour progression when compared with parametric statistical modelling using logistic regression, with SVM achieving the highest performance. The strongest predictor of tumour progression was the extent of surgical resection, with patient age, tumour volume, and the use of radiotherapy also showing influence. No features showed an association with tumour recurrence following a complete resection. In conclusion, this study demonstrates the potential of ML models in predicting post-operative outcomes for patients with NF PitNET. Future work should look to include additional, more granular, multicentre data, including incorporating imaging and operative video data.

The prediction of the risk of rockbursts in burst-prone grounds is turned into a challenging and vital mission for most underground projects that attract great interest from engineers and researchers. In this study, a hybrid technique, the artificial neural network (ANN) and artificial bee colony (ABC), neuro-bee model, was considered to create the sophisticated relationship between the risk of rockbursts in burst-prone grounds and its influencing factors. The establishment and validation of ANN models were implemented via a data set extracted from previous works, and the database covers 246 reliable rockburst cases. Six influencing factors were selected as input variables. Five-fold cross validation were adopted to tune hyper-parameters of ABC-ANN models, and the performance of ANN models was evaluated by correlation coefficient (R2) and root mean square error (RMSE). Observational experiment results indicated that the ABC-ANN algorithm can be utilized as an effective tool for predicting the risk of rockbursts in burst-prone grounds. The R2 and RMSE values between the predicted and actual rockburst values were 0.9656 and 0.1281, respectively. Sensitivity analyses implemented by the response surface method revealed that the maximum tangential stress of the cavern wall and the elastic strain index parameters have a greater effects on rockburst compared with other input parameters. As a result, the proposed hybrid method outperforms the other models for rockburst prediction in terms of the prediction accuracy and the generalization capability.

Parkinson’s disease (PD) is a neurodegenerative disease cause by a deficiency of dopamine. Investigators have identified the voice as the underlying symptom of PD. Advanced vocal disorder studies provide adequate treatment and support for accurate PD detection. Machine learning (ML) models have recently helped to solve problems in the classification of chronic diseases. This work aims to analyze the effect of selecting features on ML efficiency on a voice-based PD detection system. It includes PD classification models of Random forest, decision Tree, neural network, logistic regression and support vector machine. The feature selection is made by RF mean-decrease in accuracy and mean-decrease in Gini techniques. Random forest kerb feature selection (RFKFS) selects only 17 features from 754 attributes. The proposed technique uses validation metrics to assess the performance of ML models. The results of the RF model with feature selection performed well among all other models with high accuracy score of 96.56% and a precision of 88.02%, a sensitivity of 98.26%, a specificity of 96.06%. The respective validation score has an Non polynomial vector (NPV) of 99.47%, a Geometric Mean (GM) of 97.15%, a Youden’s index (YI) of 94.32%, and a Matthews’s correlation method (MCC) 90.84%. The proposed model is also more robust than other models. It was also realised that using the RFKFS approach in the PD results in an effective and high-performing medical classifier.

Enhancing thermal comfort prediction in high-speed trains through machine learning and physiological signals integration

A better understanding of rockburst precursors and high stress distribution characteristics can allow for higher extraction efficiency with reduced safety concerns. Taking the rockburst that occurred on 30 January 2015 in the Sanhejian Coal Mine, Jiangsu Province, China, as an example, the mechanism of rockburst development in a roadway was analysed, and a combined method involving b values and seismic velocity tomography was used to assess the rockburst in both time and space, respectively. The results indicate that before the rockburst, b values dropped significantly from 0.829 to 0.373. Moreover, a good agreement between a significant decrease in b values and the increase of the number of strong tremors was found. Using seismic tomography, two rockburst risk areas were determined where the maximum velocity, maximum velocity anomaly and maximum velocity gradient anomaly were 6 km/s, 0.14 and 0.13, respectively. The high-velocity regions corresponded well with the rockburst zone and large seismic event distributions. The combination of b values and seismic tomography is proven to have been a promising tool for use in evaluating rockburst risk during underground coal mining.

Introduction: Decision-making for orchiectomy following testicular torsion often relies on subjective clinical evaluations.This study investigated the efficacy of machine-learning (ML) models in objectively predicting post-torsion testicular viability, aiming to maintain a parenchymal ratio over 80% compared to the contralateral testicle, irrespective of initial appearance and surgical timing.Methods: A prospective database from a single surgeon (2020-2024) covering all patients who underwent detorsion and subsequent bilateral orchidopexy was used.Followup ultrasounds were conducted at 6-12 months post-procedure.Variables included patient age (neonatal, prepubertal, post-pubertal), time from presentation to surgery, and history of torsion-detorsion events.Various ML models -regression, k-nearest neighbors, and decision trees -were assessed for accuracy and precision. Results:The decision tree model demonstrated the highest accuracy at 90.5%, followed by the regression model at 82.7% accuracy, and the k-nearest neighbors model at 81.4% accuracy.The area under the curve (AUC) for the regression model indicated adequate predictability for testicular viability.A crucial finding was the impact of the timing of surgical intervention; surgeries conducted within four hours showed a 100% viability rate (p=0.001).Age also significantly affected outcomes, with post-pubertal patients showing a higher viability rate of 57.1% (p=0.041).Detailed performance metrics, such as precision, recall, and F1-scores for each model, further validate the predictive capacity of these ML models (Figure 1, Tables 1,2).Conclusions: Preliminary results suggest that ML models are viable tools for predicting testicular viability post-torsion, significantly enhancing clinical decisionmaking by providing an objective basis for potentially preserving testicles.Further studies with larger datasets are necessary to confirm and refine these predictions.

In recent years, a significant amount of research has focused on analyzing the effectiveness of machine learning (ML) models for malware detection. These approaches have ranged from methods such as decision trees and clustering to more complex approaches like support vector machine (SVM) and deep neural networks. In particular, neural networks have proven to be very effective in detecting complex and advanced malware. This, however, comes with a caveat. Neural networks are notoriously complex. Therefore, the decisions that they make are often just accepted without questioning why the model made that specific decision. The black box characteristic of neural networks has challenged researchers to explore methods to explain black-box models such as SVM and neural networks and their decision-making process. Transparency and explainability give the experts and malware analysts assurance and trustworthiness about the ML models’ decisions. In addition, it helps in generating comprehensive reports that can be used to enhance cyber threat intelligence sharing. As such, this much needed analysis drives our work in this paper to explore the explainability and interpretability of ML models in the field of online malware detection. In this paper, we used the Shapley Additive exPlanations (SHAP) explainability technique to achieve efficient performance in interpreting the outcome of different ML models such as SVM Linear, SVM-RBF (Radial Basis Function), Random Forest (RF), Feed-Forward Neural Net (FFNN), and Convolutional Neural Network (CNN) models trained on an online malware dataset. To explain the output of these models, explainability techniques such as KernalSHAP, TreeSHAP, and DeepSHAP are applied to the obtained results.

Intrusion detection plays a critical role in safeguarding computer networks from malicious activities and unauthorized access. Traditional machine learning (ML) models have been widely employed for intrusion detection, but their effectiveness can be limited due to the complex and evolving nature of network attacks. An efficient Convolutional Neural Network (CNN) that utilizes one-dimensional convolutions, batch normalization, and max pooling layers is designed to capture the attack patterns in network traffic data. By leveraging the inherent feature extraction capabilities of CNNs, the proposed architecture demonstrates improved performance in identifying network intrusions compared to traditional ML models. The effectiveness of the novel CNN architecture is evaluated by conducted comprehensive experiments on two benchmark intrusion detection datasets. The designed CNN performance was compared with four other ML models such as logistic regression, k-nearest neighbors, decision trees, and random forest. The experimental results revealed that the proposed CNN architecture outperformed the traditional ML models with highest accuracy of 97.68% and 97.39% and highest F1-score of 97.68% and 97.40% for. UNSW-NB15 and CICIDS2017 datasets, respectively. The proposed CNN architecture serves as a promising alternative to traditional ML models, offering improved performance in identifying and mitigating network attacks.

Social determinants of health (SDOH) play a critical role in the onset and progression of chronic kidney disease (CKD). Despite the well-established role of SDOH, previous studies have not fully incorporated these factors in predicting CKD in Type 2 diabetes patients. To bridge this gap, this study aimed to develop and evaluate the machine learning (ML) models that incorporate SDOH to enhance CKD risk prediction in Type 2 diabetes patients. Data were obtained from the 2023 Behavioral Risk Factor Surveillance System (BRFSS), a national survey that collects comprehensive health-related data from adults across the United States. Missing data were addressed using the K-nearest neighbor imputation method, and the Synthetic Minority Oversampling Technique (SMOTE) was applied to balance class distributions. Potential predictive features were selected using correlation coefficient analysis. The dataset was partitioned into training (80%) and testing (20%) subsets, with a 3-fold cross-validation strategy applied to the training data. Seven ML models were developed for CKD risk prediction, including logistic regression (LR), decision tree (DT), K-nearest neighbor (KNN), random forest (RF), eXtreme Gradient Boosting (XGBoost), and an artificial neural network (ANN). Model performance was evaluated using multiple metrics, including the area under the receiver operating characteristic curve (AUROC), precision, recall, F1 score, accuracy, and false positive rate. The study included 19,912 Type 2 diabetes patients (weighted sample size: 818,878), among whom 2,924 (weighted 13.92%) had CKD, and 16,988 (weighted 86.08%) did not. Over half of the CKD group (50.4%) were aged 65 or older. The proportion of female patients was higher in both groups, comprising 53.8% of the CKD group and 50.5% of the non-CKD group. Among the ML models evaluated, the RF model demonstrated the highest predictive performance for CKD, with an AUROC of 0.89 (95% CI: 0.88 - 0.90), followed by the DT model (0.84, 95% CI: 0.83 - 0.85) and XGBoost (0.83, 95% CI: 0.82 - 0.84). The RF model achieved an accuracy of 0.81 (95%CI: 0.81 - 0.81), a precision of 0.79 (95%CI: 0.79 - 0.79), a recall of 0.85 (95%CI: 0.85 - 0.85), and an F1 score of 0.82 (95%CI: 0.82 - 0.82). Additionally, the RF model exhibited strong calibration, reinforcing its reliability as a predictive tool for CKD risk in individuals with Type 2 diabetes. The study findings underscore the potential of ML models, particularly the RF model, in accurately predicting CKD among individuals with Type 2 diabetes. This approach not only enhances the precision of CKD prediction but also highlights the importance of addressing social and environmental disparities in disease prevention and management. Leveraging ML models with SDOH can lead to earlier interventions, more personalized treatment plans, and improved health outcomes for vulnerable populations. The online version contains supplementary material available at 10.1007/s40200-025-01621-9.

Most log-based anomaly detectors assume logs are stable, though in reality they are often unstable due to software or environmental changes. Anomaly detection on unstable logs (ULAD) is therefore a more realistic, yet under-investigated challenge. Current approaches predominantly employ machine learning (ML) models, which often require extensive labeled data for training. To mitigate data insufficiency, we propose FlexLog , a novel hybrid approach for ULAD that combines ML models — decision tree, k-nearest neighbors, and a feedforward neural network — with a Large Language Model (Mistral) through ensemble learning. FlexLog also incorporates a cache and retrieval-augmented generation (RAG) to further enhance efficiency and effectiveness. To evaluate FlexLog , we configured four datasets for ULAD, namely ADFA-U, LOGEVOL-U, SynHDFS-U, and SYNEVOL-U. FlexLog outperforms all baselines by at least 1.2 percentage points (pp) in F1 score while using much less labeled data (62.87 pp reduction). When trained on the same amount of data as the baselines, FlexLog achieves up to a 13 pp increase in F1 score on ADFA-U across varying training dataset sizes. Additionally, FlexLog maintains inference time under one second per log sequence, making it suitable for most applications, except latency-sensitive systems. Further analysis reveals the positive impact of FlexLog ’s key components: cache, RAG, and ensemble learning.

Cancer recurrences are poorly recorded within electronic health records around the world. This hinders research into the efficacy of cancer treatments. Currently, the retrospective identification of recurrence/progression diagnosis dates is achieved by staff who manually review patients' health records. This is expensive, time-consuming, and inefficient. Machine Learning models may expedite the review of health records and facilitate the assessment of alternative cancer therapies. This paper evaluates the use of four machine learning models (random forests, conditional inference trees, decision trees, and logistic regression) in identifying proxy dates of epithelial ovarian cancer recurrence/progression from chemotherapy data, in 531 patients at Leeds Teaching Hospital Trust. The random forest achieved the highest F1 score of 0.941 (95% confidence interval 0.916-0.968) when identifying recurrence events. Both the random forest and decision tree models' classifications closely conform to chart-reviewed time to next treatment, serving as a surrogate for recurrence-free survival. Additionally, all models reached an F1 score >0.940 when identifying patients whose cancer recurred/progressed. Our models proficiently identify both proxy dates for recurrence/progression diagnoses and patients whose cancer recurred/progressed. Considering the similar performance of the random forest and decision tree, model preference should be determined by the interpretability required to assist chart review and the ease of implementation into existing architecture.