WoS İndeksli Yayınlar Koleksiyonu / WoS Indexed Publications Collection

Permanent URI for this collectionhttps://hdl.handle.net/11147/7150

Browse

Filters

20253

Settings

Author: search.filters.author.Gokalp, Osman

Search Results

Now showing 1 - 3 of 3
  • Article
    Ensemble Machine Learning Algorithms for Thermal Comfort Prediction in HVAC Systems of Smart Buildings
    (Golden Light Publishing, 2025) Erdem, Merve Kuru; Gokalp, Osman; Calis, Gulben
    Predicting the thermal comfort of building occupants is of paramount importance in the operation of smart buildings, providing a data-driven approach to control Heating, Ventilation, and Air Conditioning (HVAC) systems for managing occupant thermal comfort and energy use, which aligns with modern sustainability and efficiency goals. Recently, ensemble machine learning (ML)-based thermal comfort prediction models have been proposed to provide more accurate estimation of thermal comfort; however, these efforts often lack a systematic and comprehensive evaluation across a wide range of ML models within a single study. To address this gap, this study presents a systematic comparative analysis of four ensemble ML frameworks (bagging, boosting, stacking, and voting) with six basic ML algorithms (Logistic Regression, K-Nearest Neighbors, Support Vector Machine, Decision Tree, Multilayer Perceptron, and Multinomial Na & iuml;ve Bayes) and six advanced ensemble ML algorithms (Random Forest, Rotation Forest, Extra Trees, Gradient Boosting Classifier, Histogram Gradient Boosting Classifier, and Extreme Gradient Boosting). The analysis is conducted using the widely recognized ASHRAE Global Thermal Comfort Database II, providing both 3-point and 7-point Thermal Sensation Vote (TSV) predictions. Accuracy, precision, recall and F1 metrics are used for evaluation and 10-fold cross validation is applied for further comparison. The results demonstrate the Histogram Gradient Boosting (HGB) algorithm achieved the highest F1 score (0.638) for 7-point TSV prediction whereas the Random Forest (RF) algorithm provided the highest F1 score (0.549) for 7-point TSV prediction. In practice, these findings suggest that integrating RF and HGB models into Building Management Systems or IoT-based HVAC platforms can support real-time adaptive control, helping practitioners to reduce energy use while maintaining occupant comfort.
  • Article
    Prediction of HVAC Operational Variables Using Recurrent Neural Networks for Advanced Control Strategies
    (Elsevier, 2025) Erdem, Merve Kuru; Gokalp, Osman; Calis, Gulben
    Heating, Ventilation, and Air Conditioning (HVAC) systems are major energy consumers in buildings, making accurate prediction of their operational variables essential for improving energy efficiency and occupant comfort. This study compares three recurrent neural network (RNN) architectures, namely standard RNN, Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU) for multi-input, multi-output prediction of nine HVAC variables using high-resolution data from a real office building (2018-2020). A two-stage hyperparameter optimization varying network depth, width, learning rate, and batch size was applied. Results show that a shallow architecture with a single hidden layer of 16 nodes, learning rate of 0.001, and batch size of 64 offers the best balance between accuracy and generalization. While the standard RNN showed higher errors and lower stability, both LSTM and GRU performed well. The GRU achieved the lowest average rank (4.2) and mean validation loss (0.0049) across cross-validation folds, demonstrating superior consistency and reliability, and was selected for final evaluation. On the holdout test set, GRU consistently outperformed the other models, achieving R-2 > 0.81 for seven of nine variables, with a peak R-2 of 0.99 for mean indoor air temperature and an overall average R-2 of 0.80. These results highlight GRU's advantages in accuracy, stability, and efficiency, suggesting its suitability for intelligent building control. The study's scope is limited to three RNNbased architectures with a constrained hyperparameter search; future work should extend comparisons to broader model families and datasets.
  • Conference Object
    Performance Evaluation of Filter-Based Gene Selection Methods in Cancer Classification
    (IEEE, 2025) Gokalp, Osman
    With the advances in microarray technology, gene expression levels can be measured efficiently, and this data can be used to solve important problems such as cancer classification. However, microarray data suffers from the high-dimensionality problem and requires dimensionality reduction techniques such as feature selection. This study addresses the cancer classification problem using microarray datasets and comparatively evaluates the performance of different filter-based gene (feature) selection methods. To this end, 11 microarray datasets have been evaluated using 6 different filter methods, and experimental results are presented. According to the findings, the gene selection methods used can improve classification performance by 5% to 30%. Using 5-fold cross-validation, the highest accuracy rates were achieved with 32 genes selected by the gain ratio filter for the Breast and Colon datasets, and with 8 genes selected by the information gain filter for the CNS dataset.