Decision Tree Research Articles

Machine learning-assisted investigation on nonlinear vibration analysis of bio-inspired auxetic tubes

PurposeAuxetic tubular structures with negative Poisson’s ratios have gained significant attention in biomedical applications, particularly in vascular and esophageal stents, due to their potential to reduce embolism risks. This study aims to investigate the nonlinear vibration characteristics of such structures and develop accurate predictive models using machine learning (ML) techniques.Design/methodology/approachThe governing equations of auxetic tubes are derived using Hamilton’s principle and von-Kármán’s nonlinear assumptions, while Malek-Gibson relations determine their effective mechanical properties. The mechanical behavior of polylactic acid (PLA) is experimentally analyzed through tensile testing and digital image correlation (DIC) with additional insights from scanning electron microscopy. The nonlinear vibration equations are solved via the Ritz method, and vibrational behavior is assessed using the direct displacement control approach. Predictive modeling is performed using six ML algorithms – CatBoost, decision tree, random forest, gradient boosting tree, extreme gradient boosting (XGBoost) and support vector regression (SVR) – along with an artificial neural network (ANN). Response surface methodology is employed to optimize the effects of edge supports, radius ratios and auxetic cell geometry on vibrational behavior.FindingsThe results demonstrate a strong agreement between ML/ANN predictions and the analytical Ritz method, confirming the reliability of the developed models. The analysis reveals that variations in edge supports, radius ratios and auxetic cell geometry significantly influence the vibrational response of the structures. The optimized configurations enhance the structural performance, making these auxetic tubular metastructures highly suitable for biomedical applications.Originality/valueThis study uniquely integrates analytical modeling, experimental analysis and ML-based predictive modeling to comprehensively assess and optimize the vibrational behavior of auxetic tubular metastructures. The findings provide valuable insights for the design of next-generation auxetic stents, improving their mechanical performance and expanding their potential biomedical applications.

Abstract 2507: Comparative assessment of data analysis methods to enable robust 10-second diagnosis of tissue pathologies with picosecond infrared laser mass spectrometry

Abstract A survey of clinical outcome data suggests surgical approaches of the central nervous system cancers must vary based on type of cancer, its aggressiveness, chance of recurrence and associated risks. This is usually informed by feedback from a neuropathologist on call wherein the accuracy of diagnosis depends highly on the experience and training level of the pathologist. As such, to deliver robust personalized resections, the subjectivity of intraoperative diagnosis must be reduced.Picosecond infrared laser mass spectrometry (PIRL-MS) generates a fingerprint of cancer-specific tissue lipids that can form the basis for non-subjective diagnosis in only 10 seconds. Here, a differential diagnosis model must be incorporated with finer data analytics methods to help tease out cancer-specific spectral features in complex datasets.A hierarchical differential diagnosis (decision tree) model optimizing evidence-based adjustment of the extent of resection for n=62 lymphoma (no resection), n=96 metastatic carcinomas (aggressive resection) and n=216 gliomas (maximum safe margin resection), the latter comprised of n=93 isocitrate dehydrogenase-1 (IDH1) positive and n=132 negative mutational status was created. This tree flows in a logical order based on what a neuropathologist would use for diagnosis, albeit utilized for automated non-subjective diagnosis based on 10-second PIRL-MS lipid analysis. The goal was to optimize the best machine learning method at each discrimination point from a list of principal component analysis-linear discriminant analysis (PCA-LDA), partial-least squares discriminant analysis, lasso regression, random forest, XGBoost, and support vector machine.The hierarchical approach enabled simplification of class comparisons to only those in differential diagnosis, as such, eliminating the misclassification between classes at finer branches of tree. This allowed performance metrics to be generated at each discrimination point not distinguished in simultaneous prediction of all classes using a single multiclass model.Within all nodes, PCA-LDA (a linear method) showed best performance, especially compared to ensemble approaches. Using this approach, the hierarchical method delivered diagnosis of lymphoma, glioma and metastatic tumours in node 1 with sensitivity and specificity of 87% and 93%, respectively and was able to differentiate between IDH1 mutant and WT gliomas with the sensitivity and specificity of >93%, the latter not being possible in current intraoperative diagnosis workflows due to long analysis times.The utility of this optimized hierarchical approach with PIRL-MS data is manifested in the users’ ability to adjust the diagnosis depth commensurate with desired performance criteria, enabling incorporation of the clinical ‘cost’ of total resection of one tumour class compared to another. Citation Format: Darah Vlaminck, Alexa Fiorante, Lan Anna Ye, David Munoz, Sunit Das, Gelareh Zadeh, Howard Ginsberg, Arash Zarrine-Afsar. Comparative assessment of data analysis methods to enable robust 10-second diagnosis of tissue pathologies with picosecond infrared laser mass spectrometry [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2507.

Classification of fractional-order chaotic systems using deep learning methods

Abstract Recent developments in fractional calculus reveal that fractional operators enable the emergence of new chaotic behaviors that cannot be observed in integer-order systems. When the characteristics of chaotic systems are combined with fractional calculus, the complexity and unpredictability of the system are further enhanced. This study examines two simple yet topologically similar six-term chaotic systems, namely Sprott H and Sprott K, to explore their application in deep learning within the framework of fractional calculus. Through the analysis of time series data, phase portraits, Lyapunov exponents, and bifurcation diagrams, it is demonstrated that the fractional-order Sprott H (FOS-H) and fractional-order Sprott K (FOS-K) chaotic systems exhibit chaotic behavior under specific system parameters and fractional orders. A dataset of 28,800 time series samples is generated and classified using pre-trained deep learning models, including GoogleNet, MobileNet-v2, DarkNet-19, DarkNet-53, and EfficientNet-b0, with transfer learning techniques applied. Using the SGDM optimizer with a learning rate of 0.0003, all models, except DarkNet-19, achieve high classification accuracy. To assess generalization, a 3160-sample test dataset from lower order fractional chaotic systems (0.8 for FOS-H, 0.82 for FOS-K) is introduced, where DarkNet-53 achieves 100% accuracy and GoogleNet reaches 99.77%, outperforming other models. Additionally, a comparison with classical machine learning methods (Support Vector Machines (SVM), Decision Tree, Random Forest) reveals that while SVM achieves 99.77% validation accuracy, its performance declines for lower fractional orders. Test results confirm that GoogleNet offers the best balance between accuracy and resource efficiency, making it suitable for real-time applications. These findings demonstrate that deep learning models, particularly DarkNet-53 and GoogleNet, can effectively classify chaotic time series with varying fractional orders, even when encountering previously unseen system dynamics.

Abstract 2404: Change in expression of microRNAs following limb amputation in canine osteosarcoma patients predicts survival time after chemotherapy

Abstract Osteosarcoma (OS) is the most common primary bone tumour in dogs and humans. There are an estimated 8, 000 new cases of canine OS per year in the United States compared to only 1, 000 in humans. The most common primary location for both species is the appendicular skeleton, with metastases forming mostly in the lungs and bones. Appendicular OS is not only more common in dogs, but also more aggressive, making dogs a valuable spontaneously metastasizing animal model of aggressive human OS. Median survival after amputation and adjuvant chemotherapy in dogs is 1 year, less than 30% survive 2 years, and an estimated 90% of dogs develop metastases after therapy. In contrast, humans have a 76% 5-year survival rate and 15-25% will develop metastases. Predicting clinical outcomes after treatment is difficult in both species, particularly for those without clinically detectable metastatic disease at diagnosis. MicroRNAs (miRNAs) are small, non-coding RNA that control gene expression and are often dysregulated in cancer. MiRNAs share entirely or nearly identical sequences between dogs and humans and are present in blood, making them good potential biomarkers. Recently, we identified multi-miRNA models which could reliably predict survival using pre-amputation canine OS plasma samples. Tumour cells can release miRNAs into the blood via active and passive mechanisms; therefore, with the primary tumour removed after amputation, miRNA expression levels in the blood should be altered. We hypothesize these changes in miRNA expression can be used to prognosticate canine OS patients better. MiRNA expression was measured using real-time quantitative PCR in matched pre- and post-amputation plasma samples from canine OS patients (n=25) at the Ontario Veterinary College. Change in expression was calculated for each miRNA as [(post-amputation) - (pre-amputation)] for each matched pair of samples. For each dog, survival was measured as overall survival and disease-free interval (DFI). Overall survival is the time from diagnosis to death and DFI is the time from diagnosis to clinically detectable metastasis. Dogs were censored if they died from non-OS events, were lost during follow-up, or were still alive. Prognostic ability was evaluated for individual miRNAs by identifying an optimal expression cutoff point using the survminer package in R and visualized using Kaplan-Meier curves. Decision trees were grown using the rpart package in R to evaluate whether multi-miRNA models offered improved prognostic ability compared to single miRNAs. We found that a model combining miRNA-144 and miRNA-23a, for both overall survival and DFI metrics, was able to identify 3 distinct groups which reflected survival time worse, on par, and better than the median survival time reported in literature. These models offer a simple prognostic test that may have clinical applications for veterinary and human OS patients. Citation Format: Heather Treleaven, Michael Edson, Latasha Ludwig, Alicia Viloria-Petit, R. Darren Wood, R. Ayesha Ali, Geoffrey A. Wood. Change in expression of microRNAs following limb amputation in canine osteosarcoma patients predicts survival time after chemotherapy [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2404.

Predictive power of semantic information in the use context of multiword terms for their structural disambiguation

Introduction: The structural disambiguation of English multiword terms (MWT) of three or more constituents (e.g., coastal sediment transport), often known as bracketing, involves the grouping of the dependent components so that the MWT is reduced to its basic form of modifier+head, as in coastal [sediment transport], which is a right-bracketed ternary compound. This work presents a study that explored whether the bracketing of a ternary compound, when used as an argument in a sentence, can be predicted from the semantic information encoded in that sentence. Methodology: A set of 1.694 sentences were analyzed semantically and annotated with the lexical domain of the verbs, the semantic role and category of the arguments, and the semantic relation between the arguments. These semantic variables were then analyzed statistically to determine whether they are able to predict the bracketing of a ternary compound. Results: A random forest model, with the lexical domain of the verb, and the semantic role and category of the MWT, was able to predict the bracketing of the ternary compounds used as arguments in a sample of 380 MWTs (100% F1‑score). A decision tree, with solely the semantic relation of the MWT to another argument in the same sentence, was also able to predict the bracketing of the ternary compounds in the sample (94,12% F1‑score). Discussion: Only a subset of three variables was necessary for bracketing prediction with an error free performance, whereas previous research employed a minimum of 12 variables. Conclusion: The semantic information in a sentence contributed substantially to compound parsing. This suggests a novel research direction in the integration of semantic variables into syntactic parsers and machine-translation applications.

Code and Data Repository for A new upper bound for the Euclidean TSP constant

This paper presents a computer-aided proof of a novel approach to constructing a new upper bound for the Euclidean Traveling Salesman Problem (TSP) Constant, leveraging numerical methods and decision trees. While the improvement achieved is modest, the approach offers a key advantage: it is primarily constrained by computer hardware, making it highly amenable to further optimization and advancements over time.

Predicting Weight Loss Success After Gastric Sleeve Surgery: A Machine Learning-Based Approach

Background/Objectives: Obesity is a global health issue, and in this context, bariatric surgery is considered the most effective treatment for severe cases. However, postoperative outcomes vary widely among individuals, driving the development of tools to predict body weight loss success. The main objective of this paper is to evaluate predictive variables for successful weight loss one year after Sleeve bariatric surgery, defining success as a weight loss exceeding 30%. Methods: A dataset of 94 cases was included in this study. Data were collected between 2013 and 2018 from the Nutrition Section of the Endocrinology and Nutrition Department in the Eastern Area of Valladolid, Spain. Machine learning algorithms applied included Random Forest, Multilayer Perceptron, XGBoost, Decision Tree, Logistic Regression, and Support Vector Machines (SVMs). Results: The SVM model demonstrated the best performance, attaining an accuracy of 88% and an area under the curve (AUC) of 0.76 with a 95% CI between 0.5238 and 0.9658. The main predictive variables identified were potassium (K), folic acid, alkaline phosphatase (ALP), height, transferrin, weight, body mass index (BMI), triglyceride (Tg), Beck Depression Test score, and insulin levels. Conclusions: In conclusion, this study highlights the potential of machine learning models, particularly Support Vector Machines (SVMs), in predicting successful weight loss after Sleeve bariatric surgery. The key predictive variables identified include biochemical markers, anthropometric measures, and psychological factors, emphasizing the multifactorial nature of postoperative weight loss outcomes.

Optimizing Hotel Efficiency Through Predictive Analysis of Reservation Cancellations

One of the most essential in hotel operational efficiency is to enhance profitability and customer satisfaction. Handling reservation cancellations is a vital part corresponding with the operational efficiency, which can have an effect on revenue management, resource allocation, and even overall service quality. By identifying underlying factors that contribute to cancellations and developing predictive models to alleviate their impact, this research explores how analyzing cancellation patterns can improve hotel efficiency. Hotels can have a prediction of cancellations more accurately, and furthermore, adjusting booking policies, and implement dynamic pricing strategies by analyzing historical data and applying classification models like Logistic Regression, Decision Tree and Random Forest. The function of flexible booking options, customer segmentation, and targeted communication to reduce last-minute cancellations by analyzing classification models is highlighted in the research. Ultimately, this method improves resource utilization, minimizes revenue loss, and improves the overall customers experience.

Abstract 7432: Transformer-based glioma patient recurrence prediction model infers novel genetic markers

Abstract Approximately 90% of glioma patients experience recurrence after initial treatment, leading to poor prognosis and short survival time. While established genetic markers, such as IDH mutation and 1p/19q codeletion, are linked to glioma prognosis, there is a critical need to identify novel targets associated with recurrence to overcome therapeutic challenges. In this study, we developed a transformer-based deep learning model to predict glioma recurrence by integrating multi-omics data, including mutations and copy number variations, from the Glioma Longitudinal Analysis (GLASS) Consortium. Our model outperformed traditional machine learning approaches, including support vector machines, logistic regression, and decision trees, achieving the best performance when utilizing mutational features alone, while copy number variations were less informative. Feature importance analysis identified key contributors to glioma recurrence, including frequently mutated genes such as TP53, ATRX, and PTEN, as well as novel markers like MUC16, TTN, and CIC. Our approach represents an early application of state-of-the-art deep learning to dissect the genomic drivers of glioma recurrence, offering valuable insights for developing targeted therapies to improve patient outcomes. Citation Format: Ko-Hong Lin, Kai Zhang, Dung-Fang Lee, Xiaoqian Jiang. Transformer-based glioma patient recurrence prediction model infers novel genetic markers [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 7432.

The Mechanism of Employee Characteristics on Promotion: Building the Predictive Machine Learning Model

Purpose: This study developed a predictive model based on employee characteristics, analyzes multidimensional data to identify predictive factors of promotion outcomes, and provides data-driven insights for organizations to improve talent management. Methods: Decision tree, random forest, and neural network models were built on the existing dataset (N=54,808). Then, the models predictive efficacies were evaluated to select the optimal model using several methods. Results: The accuracy of decision tree was 85.7%. The accuracy of artificial neural network and random forest model was 93%. For promoted class, random forest had higher precision, while neural network performed slightly better in recall and F1 score. For non-promoted class, both models perform almost identically in precision, recall, and F1 score. Conclusion: Random forest and neural network had good predictive efficacy for employee promotion. If avoiding false promotions is more important (i.e., precision matters more), random forest is the better model. If capturing more promotions is critical (i.e., recall matters more), even at the cost of some false positives, then the neural network is slightly better.

Application of Decision Tree Model in Personal Credit Scoring and Its Fairness Optimization

Credit scoring is an essential tool for assessing borrower risk in the financial industry. Traditional credit scoring methods often fail to maintain high accuracy and flexibility in the face of complex financial data. In contrast, decision tree modeling is an effective tool for improving the accuracy of credit scoring due to its simplicity and ability to handle complex data. However, decision tree models may also amplify biases in the data, leading to unfair treatment of certain groups in scoring. This paper explores several fairness optimization methods, such as data preprocessing, fairness constraints in model training, and post-processing adjustments to address this issue.It proposes a framework for balancing fairness, accuracy, and efficiency. Although these methods reduce bias to a certain extent, none of them deal with the issue of how to ensure fairness with high precision, so future research should focus on developing optimization algorithms adapted to dynamic environments, balancing multiple objectives, and improving fairness assessment criteria to achieve a fairer and more efficient credit scoring model that will promote the continued development of the financial industry and contribute to the sustainable development of society.

WOB and hook load prediction in drilling based on GA-RF

Abstract Weight on bit (WOB) and hook load are important parameters in the drilling process, and accurate application of WOB and hook load can improve drilling efficiency. The traditional mathematical model needs to consider many relevant factors, the model is difficult to establish, and the prediction accuracy of the model is poor. In order to overcome these problems, this study established a GA-RF prediction model for WOB and hook load based on real-time drilling data of a block in Sichuan Basin. Firstly, the data is filtered and denoised, and the correlation analysis of drilling parameters is carried out. It is found that the linear correlation between the WOB and other drilling parameters was not high, while the linear correlation between the hook load and the standpipe pressure is up to 0.82. Then, a random forest (RF) optimized by genetic algorithm (GA) is established according to the data. The prediction model transmits the number of decision trees and the maximum tree depth of GA to the RF. RF model is used to deal with the nonlinear relationship of drilling parameters to WOB and hook load, and is compared with other models. The research results indicate that the GA-RF algorithm outperforms other algorithms in predicting WOB and hook load, with an R2 as high as 0.976 for predicting WOB and up to 0.998 for predicting hook load. A comparison of various prediction models reveals that the GA-RF model demonstrates high predictive accuracy for both WOB and hook load.

Estimating Heterogeneous Causal Mediation Effects with Bayesian Decision Tree Ensembles

Estimating Heterogeneous Causal Mediation Effects with Bayesian Decision Tree Ensembles

Application of interpretable machine learning algorithm to predict lymph node metastasis in cutaneous malignant melanoma

Introduction: Cutaneous malignant melanoma (CMM) is the most lethal form of skin cancer worldwide. The precise prediction of lymph node metastasis is critical for personalized treatment and improved patient outcomes. However, no prior study has employed interpretable machine learning techniques to predict lymph node metastasis in CMM. This study aimed to utilize interpretable machine learning to integrate multidimensional data from the Surveillance, Epidemiology, and End Results (SEER) database—encompassing clinical characteristics, pathological information, and biomarkers of CMM—to construct various predictive models for lymph node metastasis. Methods: We constructed six machine learning models to predict lymph node metastasis using clinical, pathological, and biomarker data from 2448 patients with CMM in the SEER database.These models comprise a support vector machine, random forest (RF), XGBoost, LightGBM, adaptive boosting, and gradient boosting decision tree. The primary influential factors were identified using Gaussian Naive Bayes and gradient boosting algorithms. Shapley additive explanations (SHAP) analysis facilitates visual interpretation in individual patients. Model performance was evaluated based on accuracy, sensitivity, specificity, Brier score, and area under the receiver operating characteristic curve (AUC). Results: The RF algorithm exhibited the highest predictive performance with an AUC of 0.897, accuracy of 0.821, sensitivity of 0.876, specificity of 0.765, and Brier score of 0.086. The primary influential variables were T stage, chemotherapy, ulceration, pretreatment lactate dehydrogenase (LDH) levels, and radiation therapy. SHAP analysis confirmed a significant association and highlighted the critical function of (LDH) as a predictive biomarker. Conclusion: This study successfully established an accurate predictive model for lymph node metastasis in patients with CMM using machine-learning techniques, offering a significant reference to aid clinician treatment decisions.

An iterative ROC procedure identifies white matter tracts diagnostic for traumatic brain injury: an exploratory analysis in U.S. Veterans.

Understanding the pathophysiology of traumatic brain injury (TBI) is crucial for effectively managing care. Diffusion tensor imaging (DTI) is an MRI technology that evaluates TBI pathology in brain white matter. However, DTI analysis generates numerous measures. Choosing between them remains an obstacle to clinical translation. In this study, we leveraged an iterative receiver operating characteristic (ROC) analysis to examine white matter tracts in a group of 380 Veterans, consisting of TBI (n = 243) and non-TBI patients (n = 137). For each participant, we obtained a whole brain tractography and extracted DTI measures from 50 tracts. The ROC analyzed these variables and produced decision trees of tracts diagnostic for TBI. We expanded our findings by applying jackknife resampling. This procedure removed potential outliers and yielded tracts not observed in the initial ROCs. Finally, we used logistic regression to confirm the tracts predicted TBI status. Our results indicate ROC can identify tracts diagnostic for TBI. We also found that groups of tracts are more predictive than any single one. These analyses show that ROC is a useful tool for exploring large, multivariate datasets and may inform the design of clinical algorithms.

Digital soil mapping of soil subgroup class information in Coimbatore district using decision tree approach

The study aimed to evaluate the effectiveness of Digital Soil Mapping (DSM) compared to traditional soil mapping methods, which can help implementing precise near-real-time smart agricultural applications. Conventional soil surveys, while informative, often lack detail and are labour-intensive. DSM addresses these limitations by integrating soil data with environmental covariates and classification algorithms. Four hundred forty soil profile data points were collected from various sources and grouped according to the USDA Soil Taxonomy at the soil subgroup level. Utilizing Landsat 8 satellite data and 33 environmental covariates, the decision tree algorithm generated 56 rules to predict soil classes. Key influencing factors identified include agro-climatic zones, physiography, mean annual minimum temperature, the green wavelength region of spectral data, rainfall, and geology. The model was trained on 348 data points and validated on 92 data points, achieving a classification accuracy of 79.35% and a Kappa coefficient of 0.78, indicating high reliability. The study concludes that DSM is a viable alternative to conventional soil mapping methods, primarily using decision tree algorithms. It demonstrates that the accuracy of DSM can be significantly enhanced by incorporating a larger number of soil profile observations and relevant environmental covariates. The expert system approach provides a more detailed and up-to-date understanding of soil distribution, crucial for agricultural planning and natural resource management in the Coimbatore district, Western Tamil Nadu.

DIET AND LIFESTYLE ADVISOR

Since lifestyle diseases such as diabetes, heart disease, hypertension, and obesity are on the rise worldwide, active, personalized healthcare is essential. Traditional dietary recommendations do not address variation in health status, taste, and culture and therefore are less effective. To counter this, the intersection of artificial intelligence (AI) and machine learning (ML) with healthcare has opened up new opportunities for personalized nutrition and the prevention of disease. This work suggests an AI-powered Diet and Lifestyle Advisor able to generate tailored dietary and lifestyle recommendations from user-specific data—such as underlying medical conditions, disease history, allergies, and daily routines. The system employs advanced machine learning algorithms (such as decision trees, neural networks, collaborative filtering) to digest health data and generate dynamic, evidence-driven meal plans optimized according to the specific needs of every user. The system further comprises a straightforward web interface, secure database administration, and eventual integration with wearable technology and electronic health records (EHRs). Leaning on real-time feedback and ongoing learning mechanisms, the advisor is able to promote user engagement, improve diet compliance, and facilitate long-term wellness. Overall, this research showcases the capability of AI to fill the gap between conventional nutrition counselling and tailored care, facilitating data-driven disease prevention and healthier living.

Explainable machine learning for predicting lung metastasis of colorectal cancer

Patients with lung metastasis of colorectal cancer typically have a poor prognosis. Therefore, establishing an effective screening and diagnosis model is paramount. Our study seeks to construct and verify a predictive model utilizing machine learning (ML) that can evaluate the risk of lung metastasis with newly diagnosed colorectal cancer (CRC) using Shapley Additive exPlanations (SHAP). Using the Surveillance, Epidemiology, and End Results database, 39,674 were extracted for model development, all of whom had been pathologically diagnosed with CRC. The data spans from 2010 to 2015. Our study has constructed seven ML algorithms based on the data mentioned above, including Random Forest (RF), Decision Tree, Support Vector Machine, Naive Bayes, K-Nearest Neighbor, eXtreme Gradient Boosting, and Gradient Boosting Machine. We selected the best algorithm and visualized it using SHAP. We conducted a validation of the model utilizing data from a Chinese hospital to assess its practicality. Based on this, we have constructed an open web calculator. 39,674 patient data were included in our study, among whom 1369 (3.5%) presented with distant lung metastasis. The Random Forest (RF) algorithm demonstrated the highest predictive capability within the internal test set (AUC of 0.980, AUPR of 0.941). Furthermore, the random forest algorithm also exhibited excellent performance in external validation sets. Meanwhile, we have also established a web calculator (http://121.43.117.60:8003/). The RF algorithm has demonstrated excellent predictive performance. It can assist clinicians in devising more personalized treatment plans.

Predicting and Preventing School Dropout with Business Intelligence: Insights from a Systematic Review

School dropout in higher education remains a significant global challenge with profound socioeconomic consequences. To address this complex issue, educational institutions increasingly rely on business intelligence (BI) and related predictive analytics, such as machine learning and data mining techniques. This systematic review critically examines the application of BI and predictive analytics for analyzing and preventing student dropout, synthesizing evidence from 230 studies published globally between 1996 and 2025. We collected literature from the Google Scholar and Scopus databases using a comprehensive search strategy, incorporating keywords such as “business intelligence”, “machine learning”, and “big data”. The results highlight a wide range of predictive tools and methodologies, notably data visualization platforms (e.g., Power BI) and algorithms like decision trees, Random Forest, and logistic regression, demonstrating effectiveness in identifying dropout patterns and at-risk students. Common predictive variables included personal, socioeconomic, academic, institutional, and engagement-related factors, reflecting dropout’s multifaceted nature. Critical challenges identified include data privacy regulations (e.g., GDPR and FERPA), limited data integration capabilities, interpretability of advanced models, ethical considerations, and educators’ capacity to leverage BI effectively. Despite these challenges, BI applications significantly enhance institutions’ ability to predict dropout accurately and implement timely, targeted interventions. This review emphasizes the need for ongoing research on integrating ethical AI-driven analytics and scaling BI solutions across diverse educational contexts to reduce dropout rates effectively and sustainably.

Comparative Analysis of Machine Learning and Deep Learning Models for Lung Cancer Prediction Based on Symptomatic and Lifestyle Features

Lung cancer remains a leading cause of global mortality, with early detection being critical for improving the patient survival rates. However, applying machine learning and deep learning effectively for lung cancer prediction using symptomatic and lifestyle data requires the careful consideration of feature selection and model optimization, which is not consistently addressed in the existing research. This research addresses this gap by systematically evaluating and comparing the predictive efficacy of several machine learning and deep learning models, employing rigorous data preprocessing, including feature selection with Pearson’s correlation, outlier removal, and normalization, on a patient symptom and lifestyle factor dataset from Kaggle. Machine learning classifiers, including Decision Trees, K-Nearest Neighbors, Random Forest, Naïve Bayes, AdaBoost, Logistic Regression, and Support Vector Machines, were implemented using Weka simultaneously with neural network models with 1, 2, and 3 hidden layers, which were developed in Python within a Jupyter Notebook environment. The model performance was assessed using K-fold cross-validation and 80/20 train/test splitting. The results highlight the importance of feature selection for enhancing the model accuracy and demonstrate that the single-hidden-layer neural network, trained for 800 epochs, achieved a prediction accuracy of 92.86%, outperforming the machine learning models. This study contributes to developing more effective computational methods for early lung cancer detection, ultimately supporting improved patient outcomes.