Selahattin Barış Çelebi

Short-term global horizontal irradiance (GHI) forecasting is essential for solar grid integration in hot arid sites due to atmospheric variability and dust aerosols. This study proposes a hybrid deep learning (DL) pipeline that combines Discrete Wavelet Transform (DWT), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) layers for 15-min GHI forecasting at a site in Riyadh, Saudi Arabia. Wavelet decomposition and model hyperparameters are jointly optimized using Optuna’s Bayesian search across 20 trials, and SHAP-based feature selection retains only the most informative predictors. Uncertainty is quantified through Monte Carlo dropout across 100 stochastic forward passes with a dropout rate of 0.2. The model is evaluated on 105,312 meteorological records using five-fold time-series cross-validation. Across all observations, the model achieves a mean RMSE of 6.62 ± 2.22 W/m ² and an R ² of 0.98 ± 0.01. Daytime-filtered evaluation (GHI > 50 W/m ² ) yields a mean RMSE of 8.28 ± 1.45 W/m ² and an R ² of 0.961 ± 0.015, providing a more conservative estimate of forecasting accuracy during operationally relevant periods. The full pipeline reduces RMSE by 28–40% compared to ablated model variants and achieves a 47% reduction compared to the persistence baseline. The proposed pipeline delivers interpretable solar irradiance forecasting with calibrated uncertainty estimates suited for grid management applications at arid-climate sites.

Generalized Anxiety Disorder (GAD) and Separation Anxiety Disorder (SAD) are common in preschool children but are often not recognized in routine care. The Preschool Age Psychiatric Assessment (PAPA) provides detailed diagnostic information, yet its use in everyday clinical practice is limited because it is time-consuming and requires trained interviewers. In this study, we developed explainable machine learning models to classify GAD and SAD using the publicly available Preschool Anxiety Study cohort (n = 917; ages 2–5; 76 raw PAPA items). We trained Logistic Regression, Random Forest, XGBoost, and LightGBM models with SMOTE-Tomek resampling applied only to the training data and performed Optuna TPE-based hyperparameter optimization for XGBoost and LightGBM. Model calibration was evaluated using the Brier score, Expected Calibration Error, and the Hosmer–Lemeshow test, while SHAP TreeExplainer was used to interpret model decisions at both global and individual levels. LightGBM achieved the highest cross-validation AUC values (GAD: 0.9928; SAD: 0.9897), and XGBoost showed strong and well-balanced performance on the independent test set (GAD AUC = 0.9865, MCC = 0.9055; SAD AUC = 0.9948, MCC = 0.9057). LightGBM was the only well-calibrated model for GAD, whereas both gradient boosting models showed very good calibration for SAD. A reduced 10-item model provided a practical screening prototype for GAD in primary care by offering higher sensitivity with an AUC very close to the full model. However, external validation in independent samples is still needed to confirm these findings.

A considerable portion of worldwide energy usage and carbon emissions comes from buildings. Accurate short-term electricity load forecasting is critical to optimize building energy management systems. In this study, the PatchTST model was evaluated for short-term electricity load forecasting at the building scale. This forecasting problem was addressed using 50 buildings from the public Building Data Genome Project 2 dataset. Experiments were conducted using a strict temporal split strategy, where models generated 24-hour forward predictions from 336 hours of historical data. Temporal features, including hour, day, month, and weekend indicators, were integrated into the PatchTST model. The model was compared against Persistence, Long Short-Term Memory (LSTM), and DLinear baselines. On the test set, the PatchTST model outperformed all baselines, achieving a mean absolute error (MAE) of 7.24 kWh, a root mean square error (RMSE) of 13.85 kWh, an R² of 0.9894, and a symmetric mean absolute percentage error (MAPE) of 9.42%. Per-horizon analyses showed that PatchTST maintained stable performance across the forecast horizon. A building-level Wilcoxon signed-rank test indicated that the superiority of PatchTST over all baselines was statistically significant (p < 0.001) with large effect sizes These findings suggest that patch-based Transformer architectures provide high accuracy, statistical reliability for short-term electricity load forecasting.

Short-term wind power forecasting supports grid stability despite wind’s intermittency, but deep learning (DL) models often require excessive computation, hindering real-time deployment. This study compares three classical regression models, linear regression (LR), k-nearest neighbors (KNNs), and support vector regression (SVR) using 10-minute SCADA data. The data retained raw variability, with only negative power values set to zero. Hyperparameters of KNN and SVR were optimized using GridSearchCV with time series cross-validation. Optimized SVR achieved superior performance (MAE = 183.93 kW, RMSE = 387.64 kW, R ² = 0.91, MAPE = 144.88%, MdAE = 84.44 kW), outperforming KNN (R ² = 0.79) and LR (R ² = 0.91 but higher errors). Scatter plots and residuals highlighted SVR’s ability to handle nonlinearities. Results suggest that SVR is a viable candidate SVR for low-resource, real-time forecasting in renewables.

Objective: This study aims to construct a high-precision decision support model by integrating explainable artificial intelligence (XAI) techniques utilizing urinary and plasma biomarkers. Methods: We used an open-access dataset published by Debernardi et al. (2020) containing the biomarkers Lymphatic Vessel Endothelial Hyaluronan Receptor 1 (LYVE1), Regenerating Family Member 1 Beta (REG1B), Trefoil Factor 1 (TFF1), and plasma CA19-9. The preprocessing phase involved missing value imputation, Z-score normalization, and feature engineering. An autoencoder (AE)-based unsupervised learning framework was employed for dimensionality reduction. Classification was performed using an XGBoost algorithm optimized via the Optuna framework. Class imbalance was addressed through the Synthetic Minority Over-sampling Technique (SMOTE). Model interpretability was ensured using SHapley Additive Explanations (SHAP). Results: The proposed Autoencoder–XGBoost model optimized with Optuna outperformed conventional methods, achieving an accuracy of 95.8%, 95% precision, 93% recall, 93% F1-score, and an AUC of 0.984. SHAP analysis identified plasma CA19-9, LYVE1, creatinine, and age as the most influential predictors contributing to model decisions. Conclusion: The developed XAI framework offers high diagnostic accuracy and transparent decision logic for the early detection of PDAC. By leveraging the clinical potential of urinary biomarkers, the model demonstrates strong applicability for integration into screening and risk stratification modules of clinical decision support systems.

Quality control in traditional pottery production is a labor-intensive process that relies heavily on manual visual inspection to detect structural defects. To address this, the present study proposes an automated framework combining image processing techniques with Artificial Neural Networks (ANN) to detect and classify cracks in handmade earthenware jugs. Images acquired via smartphone were preprocessed using Otsu’s thresholding and median filtering to effectively isolate defect regions and suppress noise. A dataset of 189 samples was utilized to train and test the ANN model. The proposed model achieved a classification accuracy of 92.98%. Confusion matrix analysis confirmed robust performance, demonstrating high capability in distinguishing between intact and defective samples with no significant bias toward either class. This approach offers potential for modernizing pottery production, enhancing efficiency, and ensuring quality.

Aims: The aim of this study was to develop an OWL-based ontology that accurately represents the structure of a university vocational school at the individual level, with semantic depth and logical coherence. Methods: In this study, an OWL-based ontology representing the institutional structure of a vocational school affiliated with a university was developed. First, a domain analysis was conducted; core entities such as courses, programs, academic and administrative staff, and students, as well as their relationships, were identified. Then, this conceptual structure was formalized in OWL using the Protégé 5.6.6 editor. Object properties (such as teaches, learns) and data properties (such as studentNumber, department) were defined to model inter-individual relationships and attributes. By adding individuals that represent real-world scenarios, the instance-level applicability of the ontology was ensured. Logical consistency was validated with the FaCT++ 1.6.5 reasoner. This approach supported both the conceptual integrity of the ontology and its potential for integration into educational management systems. Results: The developed ontology effectively modeled the components of the vocational school, such as courses, personnel, programs, and learning environments, at the class, property, and individual levels. In the validation conducted with the FaCT++ 1.6.5 reasoner, logical consistency was ensured, and the potential for integration with real-world applications was indicated through individuals defined at the instance level. The ontology provides a robust foundation for semantic querying and information extraction. Conclusion: The developed ontology provides semantic consistency and real-world applicability in educational systems