Learning Analytics to Uncover Ethnic Bias in Educational Texts: An Ensemble Learning Approach

Examining the electrical discharge machining (EDM) process is challenging in manufacturing technology due to the complexity of the physical events that take place in the gaps between electrodes. In this study, we examined the EDM process in detail and developed multiple machine learning (ML) models to describe the relationship between the EDM independent (process parameters) and dependent (responses) variables. The collected experimental data was used to train the machine learning models. According to the results, the GPR model outperformed other ML models across different materials, with average RMSE values of 0.9234 and 3.0216 for the material removal rate (MRR) and surface roughness (Sa), respectively, highlighting the effectiveness of ML tools at modeling complex machining processes, such as EDM. In addition, as a practical implication, this study opens the door to employing the developed ML models to predict the EDM process performance.

Recognition and prediction of individual thermal comfort requirement based on local skin temperature

Autism spectrum disorder (ASD) is a multifaceted neurodevelopmental condition that affects millions of individuals globally. This study leverages the publicly available UCI Machine Learning ASD Screening Dataset, which contains behavioral questionnaire responses from children aged 4-11 years, to develop and compare multiple machine learning (ML) models for early ASD diagnosis. This study applies SHAP (SHapley Additive exPlanations) to interpret and analyze the feature importance of various attributes in the dataset. Advanced ML models like Extreme Gradient Boosting (XGBoost), Random Forest (RF), Extra Trees Classifier (ETC), Support Vector Machines (SVM), and Multilayer Perceptrons (MLPs) were implemented to evaluate and compare their performance in predicting early ASD. The developed ASD model underwent a rigorous evaluation through comprehensive performance metrics, including accuracy, precision, recall, F1-score, specificity, and Matthews correlation coefficient. Ranking-based metrics, such as area under the receiver operating characteristic curve (AUROC), provided additional insights into classification robustness, while 10-fold cross-validation ensured reliability and generalizability. Results indicate that XGBoost achieved the highest diagnostic performance (accuracy = 97.8%, AUROC = 0.99), followed closely by MLP, with RF and ETC offering balanced performance and interpretability. SHAP analysis identified key behavioral features influencing model predictions, enabling explainable AI-driven insights. These findings suggest that ML-based tools can support clinicians in identifying ASD earlier, allowing for timelier interventions in disability services, particularly in underserved populations. While results are promising, the use of secondary data and the lack of clinical validation remain limitations, highlighting the need for real-world pilot testing. These findings underscore the transformative potential of ML techniques in enhancing ASD diagnostics. Despite their promise, the results are constrained by reliance on secondary data and the absence of clinical validation, underscoring the need for real-world pilot testing. Nonetheless, incorporating these models into clinical practice could accelerate early detection, facilitate personalized interventions, and optimize resource allocation for improved ASD management outcomes.

Purpose: Accurate survival prediction is essential for optimizing the treatment planning in patients with castration-resistant prostate cancer (CRPC). However, the traditional statistical models often underperform due to limited variable inclusion and an inability to account for complex, multidimensional data interactions. Methods: We retrospectively collected 46 clinical, laboratory, and pathological variables from 801 patients with CRPC, covering the disease course from the initial disease diagnosis to CRPC progression. Multiple machine learning (ML) models, including random survival forests (RSFs), XGBoost, LightGBM, and logistic regression, were developed to predict cancer-specific mortality (CSM), overall mortality (OM), and 2- and 3-year survival status. The dataset was split into training and test cohorts (80:20), with 10-fold cross-validation. The performance was assessed using the C-index for regression models and the AUC, accuracy, precision, recall, and F1-score for classification models. Model interpretability was assessed using SHapley Additive exPlanations (SHAP). Results: Over a median follow-up of 24 months, 70.6% of patients experienced CSM. RSFs achieved the highest C-index in the test set for both CSM (0.772) and OM (0.771). For classification tasks, RSFs demonstrated a superior performance in predicting 2-year survival, while XGBoost yielded the highest F1-score for 3-year survival. The SHAP analysis identified time to first-line CRPC treatment and hemoglobin and alkaline phosphatase levels as key predictors of survival outcomes. Conclusion: The RSF and XGBoost ML models demonstrated a superior performance over that of traditional statistical methods in predicting survival in CRPC. These models offer accurate and interpretable prognostic tools that may inform personalized treatment strategies. External validation and the integration of emerging therapies are warranted for broader clinical applicability.

An ensemble approach for enhanced Day-Ahead price forecasting in electricity markets

The growing adoption of smart education systems has resulted in extensive digital records of student learning activities, creating new possibilities for analysing academic progress through learning analytics. These data-driven approaches allow educational institutions to move beyond traditional evaluation methods and gain deeper insights into student performance patterns. This paper examines how learning analytics can be applied to support both the prediction of academic outcomes and the improvement of student performance in digitally enabled learning environments. The analysis is based on educational datasets containing information related to learner demographics, prior academic results, attendance behaviour, and engagement with online learning platforms. Statistical methods are used to explore relationships among learning variables, while predictive techniques based on regression and classification models are employed to estimate academic performance and detect students who may require additional academic support. Model effectiveness is assessed using standard evaluation measures to ensure dependable performance. The outcomes of the analysis demonstrate that learning analytics provides actionable insights into student academic behaviour and enables early identification of performance-related risks. These insights can assist educators in implementing timely, focused, and personalised academic interventions. The study highlights the practical value of integrating learning analytics into smart education systems to strengthen instructional planning and institutional decision-making. By promoting data-informed educational practices, learning analytics contributes to improved academic achievement and supports the ongoing digital transformation of modern education systems.

A systemic analysis of gender equality trends in the field of digital education and information technology in Ukraine for the period 2019–2025 was carried out using modern learning analytics methods. The research is based on representative statistical sources – data from the Unified State Electronic Database on Education (EDEBO), UNESCO and OECD reports, and Coursera Global Skills Report analytics. In the course of the work, it was established that the share of women among IT students in Ukrainian universities demonstrates a steady gradual growth: from approximately 18% in 2019 to 25–27% according to the results of the 2024–2025 admission campaigns. Despite the positive dynamics, it was found that structural imbalances still persist both at the level of STEM fields in general and in individual segments of the labor market, which requires the introduction of new monitoring tools. Within the framework of the study, the potential of using learning analytics technologies in combination with the concept of explainable artificial intelligence (Explainable AI) and specialized fairness metrics, such as Demographic Parity and Equalized Odds, is analyzed in detail. These tools are used to identify and eliminate hidden algorithmic bias in automated educational systems and online learning platforms. It is shown that the integration of learning analytics into state and corporate monitoring systems allows not only to quantitatively assess the gender equality index, but also to identify latent factors that affect the level of access to technology and individual academic performance of students. Particular attention is paid to the substantiation of the need to create a national strategic framework: Gender-Aware Learning Analytics Framework (GALAF). The proposed architecture is aimed at ensuring the ethical use of big data and guaranteeing the transparency of educational algorithms, which is critical for the democratization of digital education. The obtained results confirm that the combination of high-tech data analysis methods with socio-analytical approaches is the most effective tool for identifying, visualizing, and systematically reducing gender imbalances in the field of training highly qualified information technology specialists in Ukraine. The formulated conclusions can be used to develop policies for the digital transformation of education and programs to support gender inclusion in the IT sector.

BackgroundOsteogenesis imperfecta (OI) is a genetic disorder characterized by low bone mass, bone fragility and short stature. There is a significant gap in knowledge regarding the growth patterns across different types of OI, and the prediction of height in individuals with OI was not adequately addressed. In this study, we described the growth patterns and predicted the height of individuals with OI employing multiple machine learning (ML) models. Accurate height prediction enables effective monitoring and facilitates the development of personalized intervention plans for managing OI.MethodThis study included cross-sectional data for 323 participants with OI, and the median height Z-score for OI types I, III and IV were − 0.62 (-5.93 ~ 3.24), -3.97 (-10.44 ~ -0.02) and − 1.64 (-6.67 ~ 2.44), respectively. Based on the cross-sectional data of participants, the height curves across different gender and OI types were plotted and compared. Subsequently, feature selection techniques, specifically the filter and wrapper methods, were employed to identify predictive factors for the height of participants. Finally, multiple machine learning (ML) models were constructed for height prediction, and the performance of each model was systematically evaluated.ResultsThe analysis of height curves revealed that male with OI are significantly taller than female with OI from the age of 14 (p = 0.045), individuals with OI type III are statistically shorter than those with OI types I and IV starting from 3 years old (p = 0.006), and those with OI type IV are statistically shorter than those with OI type I from the age of 10 (p = 0.028). The application of filter and wrapper methods identified gender (p = 0.001), age (p < 0.001), Sillence types (p = 0.007), weight Z-score (p < 0.001) and aBMD Z-score (p = 0.021) as significant predictive factors for height. The optimal performance of predictive models was registered by gradient boosting classifier (GB) (bias = 5.783, accuracy = 92.59%, R2 = 0.828), random forest (RF) (bias = 6.155, accuracy = 90.12%, R2 = 0.788), ensemble machine learning (EML) (bias = 6.250, accuracy = 91.36%, R2 = 0.825) and deep neuron networks (DNNs) (bias = 6.223, accuracy = 90.12%, R2 = 0.821).ConclusionThis study analyzed a large cohort of individuals with OI and provided detailed height patterns across different gender and OI types that are crucial for assessing overall growth. Gender, age, Sillence types, weight Z-score and aBMD Z-score were identified as predictive factors for height. The predictive models of GB, RF, EML and DNNs had higher accuracy to evaluate the height of individuals with OI. This study allows guardians and physicians to timely monitor the height parameters, and facilitate the creation of personalized intervention schedules tailored to the needs of individuals with OI.

Predicting epileptic seizures is a significantly more challenging task compared to seizure detection. However, most publicly available electroencephalography (EEG) datasets are geared towards detection because the ictal phase (main symptomatic period) is annotated. In contrast, prediction requires the availability of annotated preictal and interictal phases. To this end, we designed and developed a method called MLSPred-Bench that can be used for converting any EEG big data annotated for detection into ML-ready data suitable for prediction. We apply our methods to the existing EEG data corpus to generate 12 ML-ready benchmarks comprising data for training, validating, and testing seizure prediction models. Our strategy uses different variations of seizure prediction horizon (SPH) and the seizure occurrence period (SOP) to produce >150GB of ML-ready data. To illustrate the usefulness of the generated data, we technically validate all the benchmarks using multiple machine learning (ML) and deep learning (DL) models. We hope that the generated benchmarking data will be utilized by various computational groups for their seizure prediction model development.The work can be summarized as follows:1.Extract short preictal and interictal segments from long-duration annotated EEG montages.2.Generate a comprehensive list of ML-ready benchmarks with varying SPH and SOP.3.Technically validate the generated data with multiple ML and DL models with up-to 88.73 % validation accuracy4.Opensource code and related materials are available at https://github.com/pcdslab/MLSPred-Bench.

This paper presents a comprehensive review of the integration of Learning Analytics (LA), Artificial Intelligence (AI), and STEM education within classroom settings, aimed at enhancing educational outcomes. By examining the synergistic effects and interactions among LA, AI, and STEM disciplines, this review highlights how these technologies can collectively transform educational practices. It discusses the potential of LA and AI to personalize learning experiences, thereby improving engagement and academic success in STEM subjects. The paper also explores various case studies and success stories, illustrating practical implementations and the significant impact these technologies have made in schools. Additionally, it addresses the challenges and considerations related to the ethical use of AI and data privacy, providing insights into how educators and policymakers can navigate these issues. Overall, this review underscores the critical role of technology in shaping the future of education by fostering more adaptive and inclusive learning environments.

Robust quantification of greenhouse gas (GHG) balances in tea plantations is critical for evaluating their contribution to agricultural carbon neutrality. This study aimed to develop data-driven models to quantify soil organic carbon (SOC) sequestration and N2O emissions in Chinese tea plantations, evaluate their net GHG balance at the national scale, and assess the mitigation potential under alternative nitrogen management scenarios. Using a comprehensive national dataset, we compared multiple machine learning (ML) approaches with a conventional multiple linear regression (MLR) model to simulate N2O emissions and SOC changes in Chinese tea plantations. All ML models substantially outperformed the MLR model, with the Random Forest (RF) algorithm achieving the highest predictive accuracy. The RF models yielded R2 values of 0.68 for N2O emissions and 0.67 for SOC changes, with no significant prediction bias. Variable importance and marginal effect analyses revealed strong non-linear controls. Mineral N fertilizer input was the dominant driver of N2O emissions, followed by organic N input, soil clay content, and SOC. In contrast, SOC dynamics were primarily regulated by organic carbon inputs, tea plantation age, climate variables, and soil pH. National-scale simulations indicated an average N2O emission intensity of 9.03 kg N2O ha−1 yr−1 and a mean SOC sequestration rate of 0.88 t C ha−1 yr−1. Overall, SOC sequestration offset N2O emissions, rendering Chinese tea plantations a net GHG sink (−2525 Gg CO2-eq yr−1). Scenario analyses showed that mineral N reduction increased net GHG uptake by 1804 Gg CO2-eq, while organic fertilizer substitution achieved a substantially larger mitigation potential of 5961 Gg CO2-eq. By integrating SOC sequestration and N2O emissions within a unified modeling framework and applying machine-learning-based national-scale simulations, this study provides a more comprehensive and data-driven quantification of GHG balances in tea ecosystems, offering a scientific basis for evaluating their role in agricultural carbon neutrality strategies.

Multiple machine learning (ML) algorithms were developed using artificial intelligence, including Linear Regression (LR), Decision Tree (DT), Random Forest (RF), and K-Nearest Neighbor (KNN), to predict the yield of biodiesel production in an acid-catalyzed alcoholysis process using green seed canola oil. Catalyst loading, methanol-to-oil (M/O) molar ratio, and reaction time were considered as input parameters, while the yield of biodiesel production was selected as the output parameter. The performance of the developed ML models was assessed using evaluation metrics such as the coefficient of determination (R2) and the root mean squared error (RMSE). The R2 values obtained for LR, RF, DT, and KNN models were 0.80, 0.95, 0.97, and 0.84, respectively. Furthermore, the corresponding RMSE values for these models were 2.48, 1.51, 0.89, and 4.51, respectively. According to the results, the DT model exhibited superior accuracy and reliability for predicting biodiesel production compared to the other models. The values of the input variables to potentially yield the highest biodiesel output were identified through a systematic trial-and-error approach using the DT model. The results showed that a biodiesel yield of 88 % can be achieved with 5 wt% catalyst loading, a 22 M/O molar ratio, and a reaction time of 5 hours.

This research is concerned with the following question: How a model trained on mental state detection using stress inducer X would perform on detecting a stress stimulated by inducer Y? This is a scenario that can be faced often in real-life. In vision applications, detection in the wild usually implies a real-life setting with minimal constraints. In this research, we investigate the generalization between a classical stress inducer, namely, mental arithmetic stressor (MAE), and the stress induced by immersive technologies, particularly virtual reality (VR)-VR-induced stress. On a group of three participants, we acquired EEG recordings following their exposure to the two types of stressors mentioned above. Using multiple machine learning (ML) classifiers, we show that even basic ML models, when trained to differentiate between the MAE stressor and the non-stress state, can generalize well when tested on differentiating between VR-based stress and the non-stress state. Significant cross-detection performance is reported using accuracy, precision, recall, and F1-score. This research provides preliminary insights on the capacity of machine learning models to generalize well in the contexts of mental state detection. <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup>

This study explores the collision of suspension droplets against solid dry surfaces (substrates). It applies and compares multiple machine learning (ML) models for the classification of impact outcomes: splashing, breaking up, and rebound. A total of 372 experiments were performed with varied particle properties, liquid parameters, substrate properties, droplet velocity, and substrate inclination. The suggested markup indicators form the basis for computer vision processing of drop-wall impact videos. From 12 dimensionless features, 7 are selected to train ML classifiers. We tested 10 models, including Logistic Regression, KNN, SVM, ensemble methods (CatBoost, XGBoost, LightGBM, Random Forest, AdaBoost), and neural networks (MLP, TabNet). The gradient boosting model CatBoost showed the best results: accuracy and macro-averaged F1-score improved by more than 15% over the baseline, ROC-AUC exceeded 96%. Logistic Regression demonstrated robust generalization capabilities. The key physical finding is that greater sedimentation Stokes number, particle size, and inclination increase the likelihood of splashing and fragmentation. The models allow prediction of no-fragmentation outcomes, important for droplet-based 3D-printing and coating. To the best of our knowledge, this is the first study to use ML for the classification of suspension drop-wall impacts, including data from experiments with both heavier and lighter-than-liquid particles and inclined surfaces. The results obtained assist in controlling suspension droplet impact to avoid fragmentation.

The pharmaceutical industry is striving to develop innovative and promising tools, increasingly embracing new data-driven approaches, to understand, improve and accelerate the drug product development process. While extended release (ER) oral formulations offer a number of advantages, including maintenance of therapeutic drug levels, a reduction in dosing frequency, and minimization of side effects, achieving consistent drug release profiles remains a significant challenge. As a critical attribute for drug absorption into systemic circulation, in vitro dissolution testing represents a time-consuming and complex method for the evaluation of such formulations. The main objective of this study was to develop a model for predicting drug dissolution in the quality by design (QbD)-based development of ER oral hydrophilic matrix tablets comprising polyethylene oxide (PEO). Two main modeling approaches are conducted and compared: (i) model screening to fit and compare multiple predictive machine learning (ML) models and then deploy the best model, in this case, artificial neural networks (ANN), and (ii) functional data analysis (FDA) combined with the design of experiments (DoE) that fit a smoothing model to each dissolution curve as a continuous function. A dataset comprising 91 ER matrix tablet formulations was analyzed, with the dissolution data split into training, validation, and test sets (70%, 20%, and 10%, respectively). The results demonstrated that both ANN and functional DoE (FDOE) models achieved high similarity with the experimental dissolution profiles, as indicated by f2 values ranging from 48 to 88 for the FDOE and 52 to 88 for ANN. This work highlights the potential of integrating advanced data-driven modeling techniques into ER drug development to enhance dissolution prediction accuracy and streamline the formulation process, thus reducing time and costs.