Serdar Kırışoğlu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Serdar Kırışoğlu has a total of 11 co-authors. Here are the top co-authors by number of shared publications:
- Kurban Kotan
Department of Electrical Electronics and Computer Engineering, Duzce University, Dzce, Trkiye
5 shared publications - Mehmet Osman Devrim
Department of Information Technologies, Duzce University, Duzce, Turkiye
3 shared publications - Bayram Kotan
Department of Electrical Electronics and Computer Engineering, Duzce University, Dzce, Trkiye
2 shared publications - Abdurrahman Yakupoğlu
DÜZCE ÜNİVERSİTESİ FEN BİLİMLERİ ENSTİTÜSÜ
1 shared publication - Emin Sesen
Department of Computer Engineering, Dzce University, Dzce, Trkiye
1 shared publication - Ezgi Kara Timucin
Duzce Universitesi
1 shared publication
Latest Publications
- Open Access
- 29 September 2026
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 21 (9) , e0359577
Evaluating the Cyclical Hybrid Imputation Technique (CHIT) under MCAR, MAR, and MNAR missing data mechanisms: Evidence from health datasets
Choosing an appropriate imputation strategy for clinical datasets requires understanding which missing data mechanism is operative, yet most imputation benchmarks conflate performance across mechanisms. The present work addresses this gap by conducting the first mechanism-stratified comparison of three imputation approaches—CHIT, Multiple Imputation by Chained Equations via BayesianRidge (MICE), and Random-Forest-based Iterative Imputation—across Missing Completely at Random (MCAR), Missing at Random (MAR), and Missing Not at Random (MNAR) conditions. Three health datasets serve as experimental platforms: the Chronic Kidney Disease (CKD) dataset, the Heart Disease Dataset (HDD), and the Mice Protein Expression Dataset (MPED). Domain-knowledge-driven missingness patterns are constructed for each mechanism (Fig 2) at approximately 20–25% overall rates. Eight classifiers—KNN, Logistic Regression, SVC, Decision Tree, Random Forest, Gaussian Naïve Bayes, MLP, and a deep neural network—are trained on imputed data following GridSearchCV optimisation. Across all conditions, CHIT achieves near-perfect or perfect downstream classification, with SVC reaching up to 100% accuracy on the CKD dataset under MNAR, and 98.75% under MCAR and MAR. Competing methods degrade by up to 19.38 percentage points under MNAR, while CHIT’s accuracy remains stable. Two structural properties account for this resilience: an iterative enrichment of the regression training set as records are completed, and a within-record prioritisation of the most data-scarce features for model-based filling. Taken together, these results provide mechanism-specific guidance for imputation selection in health informatics pipelines. Statistical significance of classifier accuracy differences between CHIT and MICE was assessed using McNemar’s test (two-tailed, continuity-corrected chi-square) applied to the exact binary prediction vectors from each experiment [n_test = 80]. Ninety-five percent confidence intervals for all reported accuracy values were computed using the Wilson score method [z = 1.96]. The 42-configuration mean accuracy and standard deviation for CHIT are reported in Supplementary Table S1. Full McNemar results with confidence intervals are in Supplementary Table S2.
- 29 September 2026
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 21 (9) , e0359577
Open AccessEvaluating the Cyclical Hybrid Imputation Technique (CHIT) under MCAR, MAR, and MNAR missing data mechanisms: Evidence from health datasets
Choosing an appropriate imputation strategy for clinical datasets requires understanding which missing data mechanism is operative, yet most imputation benchmarks conflate performance across mechanisms. The present work addresses this gap by conducting the first mechanism-stratified comparison of three imputation approaches—CHIT, Multiple Imputation by Chained Equations via BayesianRidge (MICE), and Random-Forest-based Iterative Imputation—across Missing Completely at Random (MCAR), Missing at Random (MAR), and Missing Not at Random (MNAR) conditions. Three health datasets serve as experimental platforms: the Chronic Kidney Disease (CKD) dataset, the Heart Disease Dataset (HDD), and the Mice Protein Expression Dataset (MPED). Domain-knowledge-driven missingness patterns are constructed for each mechanism (Fig 2) at approximately 20–25% overall rates. Eight classifiers—KNN, Logistic Regression, SVC, Decision Tree, Random Forest, Gaussian Naïve Bayes, MLP, and a deep neural network—are trained on imputed data following GridSearchCV optimisation. Across all conditions, CHIT achieves near-perfect or perfect downstream classification, with SVC reaching up to 100% accuracy on the CKD dataset under MNAR, and 98.75% under MCAR and MAR. Competing methods degrade by up to 19.38 percentage points under MNAR, while CHIT’s accuracy remains stable. Two structural properties account for this resilience: an iterative enrichment of the regression training set as records are completed, and a within-record prioritisation of the most data-scarce features for model-based filling. Taken together, these results provide mechanism-specific guidance for imputation selection in health informatics pipelines. Statistical significance of classifier accuracy differences between CHIT and MICE was assessed using McNemar’s test (two-tailed, continuity-corrected chi-square) applied to the exact binary prediction vectors from each experiment [n_test = 80]. Ninety-five percent confidence intervals for all reported accuracy values were computed using the Wilson score method [z = 1.96]. The 42-configuration mean accuracy and standard deviation for CHIT are reported in Supplementary Table S1. Full McNemar results with confidence intervals are in Supplementary Table S2.
- Open Access
- 12 August 2026
- Published by Intelligence Science and Technology Press Inc. in Journal of Artificial Intelligence and Technology
Assessing AI Dependability in Acoustic NDT: Deep Learning Robustness for Gold Purity Verification under Industrial Noise
Acoustic resonance testing (ART) is a widely utilized nondestructive evaluation method for material characterization. However, its measurement reliability significantly degrades under harsh industrial noise, creating undefined operational risks. This study presents a systematic framework to quantify the robustness of deep learning-based acoustic sensing systems against environmental interference. A controlled synthetic noise injection protocol is employed, using white and pink (1/f)) noise as standardized proxies for broadband and low-frequency-dominant industrial interference. This protocol enables systematic analysis of measurement performance degradation across varying signal-to-noise ratios (SNRs) under simulated industrial acoustic conditions. The results define the operational boundaries for reliable measurement, identifying a critical SNR threshold of 20 dB where measurement uncertainty significantly increases. Furthermore, we demonstrate that spectral masking and the “Artificial Damping Effect” are the primary physical drivers of measurement error in noisy environments. Although the methodology is demonstrated on gold alloy benchmarks, the proposed reliability envelope and noise compensation approach are applicable to the acoustic evaluation of various homogeneous materials. These findings establish a quantitative robustness envelope and provide a reproducible benchmark for industrial deployment. Further validation under directly recorded field conditions is required before practical implementation.
- 12 August 2026
- Published by Intelligence Science and Technology Press Inc. in Journal of Artificial Intelligence and Technology
Open AccessAssessing AI Dependability in Acoustic NDT: Deep Learning Robustness for Gold Purity Verification under Industrial Noise
Acoustic resonance testing (ART) is a widely utilized nondestructive evaluation method for material characterization. However, its measurement reliability significantly degrades under harsh industrial noise, creating undefined operational risks. This study presents a systematic framework to quantify the robustness of deep learning-based acoustic sensing systems against environmental interference. A controlled synthetic noise injection protocol is employed, using white and pink (1/f)) noise as standardized proxies for broadband and low-frequency-dominant industrial interference. This protocol enables systematic analysis of measurement performance degradation across varying signal-to-noise ratios (SNRs) under simulated industrial acoustic conditions. The results define the operational boundaries for reliable measurement, identifying a critical SNR threshold of 20 dB where measurement uncertainty significantly increases. Furthermore, we demonstrate that spectral masking and the “Artificial Damping Effect” are the primary physical drivers of measurement error in noisy environments. Although the methodology is demonstrated on gold alloy benchmarks, the proposed reliability envelope and noise compensation approach are applicable to the acoustic evaluation of various homogeneous materials. These findings establish a quantitative robustness envelope and provide a reproducible benchmark for industrial deployment. Further validation under directly recorded field conditions is required before practical implementation.
- 21 January 2026
- Published by Duzce Universitesi Bilim ve Teknoloji Dergisi in Düzce Üniversitesi Bilim ve Teknoloji Dergisi
- Vol. 14 (1) , 216-224
Visual Feature Extraction and Machine Learning for Graphical Violence Detection: A Deep Learning-Free, Efficient Approach
This study proposes an efficient, deep learning-free approach for detecting graphically violent images to address the challenges of high computational costs and class imbalance in digital content moderation. Using the "Graphical Violence and Safe Images" dataset, we employed a hybrid feature extraction strategy combining color (3D Histogram), texture (Local Binary Patterns [LBP], Gray-Level Co-occurrence Matrix [GLCM]), and shape (Histogram of Oriented Gradients [HOG]) descriptors, followed by Analysis of Variance (ANOVA)-based feature selection. Among five machine learning models evaluated, XGBoost achieved the highest performance with 96.55% accuracy and an 84.38% Macro F1-Score on the test set. Furthermore, the proposed method offers a processing time of approximately 33.85 ms per image on a standard CPU. The results demonstrate that the proposed method offers a computationally efficient and interpretable alternative to deep learning for real-time applications.
- 21 January 2026
- Published by Duzce Universitesi Bilim ve Teknoloji Dergisi in Düzce Üniversitesi Bilim ve Teknoloji Dergisi
- Vol. 14 (1) , 216-224
Visual Feature Extraction and Machine Learning for Graphical Violence Detection: A Deep Learning-Free, Efficient Approach
This study proposes an efficient, deep learning-free approach for detecting graphically violent images to address the challenges of high computational costs and class imbalance in digital content moderation. Using the "Graphical Violence and Safe Images" dataset, we employed a hybrid feature extraction strategy combining color (3D Histogram), texture (Local Binary Patterns [LBP], Gray-Level Co-occurrence Matrix [GLCM]), and shape (Histogram of Oriented Gradients [HOG]) descriptors, followed by Analysis of Variance (ANOVA)-based feature selection. Among five machine learning models evaluated, XGBoost achieved the highest performance with 96.55% accuracy and an 84.38% Macro F1-Score on the test set. Furthermore, the proposed method offers a processing time of approximately 33.85 ms per image on a standard CPU. The results demonstrate that the proposed method offers a computationally efficient and interpretable alternative to deep learning for real-time applications.
- Open Access
- 14 October 2025
- Published by Springer Nature in BMC Anesthesiology
- Vol. 25 (1) , 1-11
Evaluation of anthropometric and ultrasonographic measurements with different machine learning methods in predicting difficult intubation: a prospective observational study
Difficult intubation is one of the most challenging scenarios to deal with due to increased morbidity and mortality. Machine learning systems can help predict this process in advance. This study aimed to predict whether patients had difficult intubation using machine learning programs for anthropometric and ultrasonographic measurements taken for preoperative airway assessment. Patients over 18 years of age with American Society of Anesthesiologists (ASA) scores I–III who underwent general anesthesia were included. Patients with a history of head/neck surgery, planned thyroidectomy, congenital or acquired airway anomalies morbidly obese patients with BMI > 40 or a known difficult airway were excluded. Preoperative modified mallampati test score and other anthropometric measurements (thyromental distance, neck circumference, mouth opening, sternomental distance) were recorded. Ultrasonographic measurements included the distance from skin to hyoid bone, skin to epiglottis, skin to vocal cords (anterior commissure), skin to trachea, MTT and hyomental distances in neck extension and neutral positions. The dataset was analyzed via eight different machine learning algorithms. We obtained data from 329 patients (62 difficult intubation cases). The Support Vector Machine algorithm achieved the highest performance, with an accuracy of 89.39%, a negative predictive value of 92.7%, and a positive predictive value of 72.7%. Among all evaluated parameters, the modified mallampati score, neck circumference, skin to epiglottic distance and tongue thickness were the strongest predictors of difficult intubation. The ability of individual bedside tests, which are commonly used, to predict difficult intubations is limited. Our study demonstrates that incorporating ultrasonographic measurements into a machine learning model, in addition to clinical airway assessments, improves predictive accuracy. Integrating our predictive model into a mobile app could provide a rapid and objective tool for preoperative airway assessment to identify difficult airways and improve patient safety in anesthesia settings. Prospective Observational.
- 14 October 2025
- Published by Springer Nature in BMC Anesthesiology
- Vol. 25 (1) , 1-11
Open AccessEvaluation of anthropometric and ultrasonographic measurements with different machine learning methods in predicting difficult intubation: a prospective observational study
Difficult intubation is one of the most challenging scenarios to deal with due to increased morbidity and mortality. Machine learning systems can help predict this process in advance. This study aimed to predict whether patients had difficult intubation using machine learning programs for anthropometric and ultrasonographic measurements taken for preoperative airway assessment. Patients over 18 years of age with American Society of Anesthesiologists (ASA) scores I–III who underwent general anesthesia were included. Patients with a history of head/neck surgery, planned thyroidectomy, congenital or acquired airway anomalies morbidly obese patients with BMI > 40 or a known difficult airway were excluded. Preoperative modified mallampati test score and other anthropometric measurements (thyromental distance, neck circumference, mouth opening, sternomental distance) were recorded. Ultrasonographic measurements included the distance from skin to hyoid bone, skin to epiglottis, skin to vocal cords (anterior commissure), skin to trachea, MTT and hyomental distances in neck extension and neutral positions. The dataset was analyzed via eight different machine learning algorithms. We obtained data from 329 patients (62 difficult intubation cases). The Support Vector Machine algorithm achieved the highest performance, with an accuracy of 89.39%, a negative predictive value of 92.7%, and a positive predictive value of 72.7%. Among all evaluated parameters, the modified mallampati score, neck circumference, skin to epiglottic distance and tongue thickness were the strongest predictors of difficult intubation. The ability of individual bedside tests, which are commonly used, to predict difficult intubations is limited. Our study demonstrates that incorporating ultrasonographic measurements into a machine learning model, in addition to clinical airway assessments, improves predictive accuracy. Integrating our predictive model into a mobile app could provide a rapid and objective tool for preoperative airway assessment to identify difficult airways and improve patient safety in anesthesia settings. Prospective Observational.
- 15 July 2025
- Published by Elsevier in Applied Acoustics
Determination of gold purity degrees using audio features with machine learning algorithms
- 15 July 2025
- Published by Elsevier in Applied Acoustics
Determination of gold purity degrees using audio features with machine learning algorithms
- Open Access
- 7 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 83254-83273
Detection of Economic Crises With Language Models and Comparative Analysis of Simple Time Series Analysis Models and Machine Learning Algorithms on the Stock Market
This study investigates the use of natural language processing language representation models as an early warning system for economic crises, and compares the performance of time series analysis and machine learning models in financial markets before and during the economic crises in order to select the best model. The data used in the research was collected based on the economic crises that occurred in Turkey in December 2021. The aim is to identify an economic crises period by using language representation models for economic news between August 2021 and January 2022. After identifying the economic crises period, short term (1 day), medium term (15 days) and long term (30 days) forecasts were made for the index of thirty companies with the highest trading volume (BIST30) of Borsa Istanbul between 01/01/2021 and 31/12/2021 and performance comparisons were made between the models. The aim is to develop an effective "smart, automatic crises detection and forecasting model selection application". The CHIT algorithm introduced in the study is a new missing data filling algorithm used in time series forecasting comparisons. Since the CHIT algorithm has a high impact on the model performance, this algorithm is used in the pre-processing step and comparisons are made.
- 7 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 83254-83273
Open AccessDetection of Economic Crises With Language Models and Comparative Analysis of Simple Time Series Analysis Models and Machine Learning Algorithms on the Stock Market
This study investigates the use of natural language processing language representation models as an early warning system for economic crises, and compares the performance of time series analysis and machine learning models in financial markets before and during the economic crises in order to select the best model. The data used in the research was collected based on the economic crises that occurred in Turkey in December 2021. The aim is to identify an economic crises period by using language representation models for economic news between August 2021 and January 2022. After identifying the economic crises period, short term (1 day), medium term (15 days) and long term (30 days) forecasts were made for the index of thirty companies with the highest trading volume (BIST30) of Borsa Istanbul between 01/01/2021 and 31/12/2021 and performance comparisons were made between the models. The aim is to develop an effective "smart, automatic crises detection and forecasting model selection application". The CHIT algorithm introduced in the study is a new missing data filling algorithm used in time series forecasting comparisons. Since the CHIT algorithm has a high impact on the model performance, this algorithm is used in the pre-processing step and comparisons are made.
- Open Access
- 24 February 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 1-22
Cyclical hybrid imputation technique for missing values in data sets
The problem of missing data in data sets is the most important first step to be addressed in the preprocessing phase. Because incorrect imputation of missing data increases the error in the modeling phase and reduces the prediction performance of the model. When it comes to health, it is inevitable to choose models that show a higher success rate. In cases where there is missing data, the performance of machine learning models may differ depending on the amount of data contained in the data set. The presence of missing data and this high rate affects the accuracy and reliability of analysis and modeling studies because it will affect the complete amount of data in the data set. Estimating and filling in the missing data very precisely, close to its real value, will provide a significant visible performance increase in the modeling phase, which is the next stage. After imputing the missing data with an artificial intelligence model rather than a random method, it is obvious that the accuracy of the model trained with this data is higher than the model trained with data filled with classical filling methods such as mean and mode. In this study, we propose a new algorithm that has been tested on many datasets to address the problems caused by missing data imputation in the dataset. The algorithm aims to impute missing values more effectively by using row-based and column-based imputation techniques together and cyclically. The algorithm takes into account individual missing values using column-based imputation features and the overall data structure using row-based imputation features. The proposed algorithm achieved 100% accuracy with some row and column-based imputation techniques on 3 different datasets used in the study. Higher accuracy was achieved compared to other imputation techniques.
- 24 February 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 1-22
Open AccessCyclical hybrid imputation technique for missing values in data sets
The problem of missing data in data sets is the most important first step to be addressed in the preprocessing phase. Because incorrect imputation of missing data increases the error in the modeling phase and reduces the prediction performance of the model. When it comes to health, it is inevitable to choose models that show a higher success rate. In cases where there is missing data, the performance of machine learning models may differ depending on the amount of data contained in the data set. The presence of missing data and this high rate affects the accuracy and reliability of analysis and modeling studies because it will affect the complete amount of data in the data set. Estimating and filling in the missing data very precisely, close to its real value, will provide a significant visible performance increase in the modeling phase, which is the next stage. After imputing the missing data with an artificial intelligence model rather than a random method, it is obvious that the accuracy of the model trained with this data is higher than the model trained with data filled with classical filling methods such as mean and mode. In this study, we propose a new algorithm that has been tested on many datasets to address the problems caused by missing data imputation in the dataset. The algorithm aims to impute missing values more effectively by using row-based and column-based imputation techniques together and cyclically. The algorithm takes into account individual missing values using column-based imputation features and the overall data structure using row-based imputation features. The proposed algorithm achieved 100% accuracy with some row and column-based imputation techniques on 3 different datasets used in the study. Higher accuracy was achieved compared to other imputation techniques.
- Open Access
- 13 February 2025
- Published by MDPI AG in Sustainability
- Vol. 17 (4) , 1536
Sustainable Economic Development Through Crisis Detection Using AI Techniques
Economics is based on data and indicators. Although their interpretation can be complicated, their effects can be calculated in advance. In other words, economic crises are not as complicated and unpredictable as natural disasters. If economic news, news that reflects the thoughts of society, and especially the experiences and predictions of economic experts, is semantically processed from the news texts written by economic experts, economic crises can be predicted long in advance. In addition, the frequency of news about crises in society is also effective. Events that affect society are often mentioned. This can be an indication of some economic crises. In this research, we attempted to detect the economic crises and inflation increases in Turkey in December 2021 and in Germany in September 2022 several months in advance with natural language processing (NLP) models. In the study, the daily news retrieved via RSS from the leading news channels and newspapers was first preprocessed and then the similarities were checked with NLP models. Finally, similarities and changes were analyzed in comparison with inflation data. It was found that similar changes a few months ago had a high correlation with inflation data.
- 13 February 2025
- Published by MDPI AG in Sustainability
- Vol. 17 (4) , 1536
Open AccessSustainable Economic Development Through Crisis Detection Using AI Techniques
Economics is based on data and indicators. Although their interpretation can be complicated, their effects can be calculated in advance. In other words, economic crises are not as complicated and unpredictable as natural disasters. If economic news, news that reflects the thoughts of society, and especially the experiences and predictions of economic experts, is semantically processed from the news texts written by economic experts, economic crises can be predicted long in advance. In addition, the frequency of news about crises in society is also effective. Events that affect society are often mentioned. This can be an indication of some economic crises. In this research, we attempted to detect the economic crises and inflation increases in Turkey in December 2021 and in Germany in September 2022 several months in advance with natural language processing (NLP) models. In the study, the daily news retrieved via RSS from the leading news channels and newspapers was first preprocessed and then the similarities were checked with NLP models. Finally, similarities and changes were analyzed in comparison with inflation data. It was found that similar changes a few months ago had a high correlation with inflation data.
- Open Access
- 29 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 159503-159518
GSelf-MapReduce: A Method for Enhancing Mapreduce Performance in Distributed Heterogeneous Data Centers
Big data are often stored close to the locations where they are generated, owing to the cost of data transfer. These stored data are moved to a single location for processing or processed at that location. In the literature, it is possible to find different methods for processing data in distributed data centers. In this study, we present a new method for data processing called GSelf-MapReduce. In the proposed method, shuffling is performed among heterogeneous data center (DC) that complete the data-processing process. To calculate the data processing cost of the reduce function of the DCs, a polynomial regression model was created using the data obtained in the test environment, and the coefficients obtained from this model were used in the decision process. The key/value pairs to be shuffled are distributed according to the cost of the DCs, considering their location. In addition, not all DCs are waited to finish their job for shuffling. DCs that complete their job perform shuffling among themselves. Thus, the keys are deduplicated between these DCs. The shuffling volume in the last phase and the total job completion time are reduced. The performance of the proposed method was compared with that of four different distributed data processing methods in the literature. As a result, this work generates 15% less shuffled data than the closest work.
- 29 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 159503-159518
Open AccessGSelf-MapReduce: A Method for Enhancing Mapreduce Performance in Distributed Heterogeneous Data Centers
Big data are often stored close to the locations where they are generated, owing to the cost of data transfer. These stored data are moved to a single location for processing or processed at that location. In the literature, it is possible to find different methods for processing data in distributed data centers. In this study, we present a new method for data processing called GSelf-MapReduce. In the proposed method, shuffling is performed among heterogeneous data center (DC) that complete the data-processing process. To calculate the data processing cost of the reduce function of the DCs, a polynomial regression model was created using the data obtained in the test environment, and the coefficients obtained from this model were used in the decision process. The key/value pairs to be shuffled are distributed according to the cost of the DCs, considering their location. In addition, not all DCs are waited to finish their job for shuffling. DCs that complete their job perform shuffling among themselves. Thus, the keys are deduplicated between these DCs. The shuffling volume in the last phase and the total job completion time are reduced. The performance of the proposed method was compared with that of four different distributed data processing methods in the literature. As a result, this work generates 15% less shuffled data than the closest work.
- Open Access
- 26 January 2024
- Published by Duzce Universitesi Bilim ve Teknoloji Dergisi in Düzce Üniversitesi Bilim ve Teknoloji Dergisi
- Vol. 12 (1) , 89-111
Pandemi Sürecinde Uzaktan Eğitimde Senkron, Asenkron Ve Hibrit Yapılmış Derslerde Veri Madenciliği İle Öğrenci Performans Analizi
Günümüzde üniversite öğrencilerinin eğitime ve eğitim materyallerine internetten erişim oranları oldukça artmıştır. Eğitimde internetin kullanılması ve ders materyallerine erişimin artmasına bağlı olarak analiz edilebilecek veri setinde artış meydana gelmiştir. Bu veri setlerinden bir tanesi de (planlı veya acilen) uzaktan eğitim sürecine geçen üniversitelerin, uzaktan eğitim sistemlerinde biriken ödev, sınav, proje, performans, devam notları ve benzeridir. Yeni Korona Virüs (Covid-19) pandemisinde Yüksek Öğretim Kurumu’nun (YÖK) tavsiyesi ile üniversiteler eğitimlerine uzaktan Asenkron, Senkron ve Hibrit yöntemlerini kullanarak devam etmiş, hatta sınavları uzaktan eğitim sisteminde yapmak zorunda kalmışlardır. Bu araştırmada, Kayseri Üniversitesinin Uzaktan Eğitim Uygulama ve Araştırma Merkezi (KAYUZEM) sisteminden alınmış veriler kullanılmıştır. Araştırma kapsamında 8319 işlenmiş veri bulunmaktadır. Bu veriler üzerinde Veri Madenciliği (VM) alanında kullanılan RapidMiner programının otomatik modelleme özelliği kullanılarak varsayılan algoritmalarla geleceğe yönelik tahminleme işlemi yapılmıştır. Varsayılan algoritmalar arasından en iyi sonucu veren Derin Öğrenme, Naive Bayes, Gradient Boosted Trees, Lojistik Regresyon kullanılmış ve bu otomatik modelleme de yer almayan k-En Yakın Komşu (k-NN) algoritması da çalışmaya dahil edilmiştir. Bu 5 algoritmanın parametreleri üzerinde değişiklikler yapılarak daha iyi sonuçlar elde edilmeye çalışılmıştır. Öğrenci başarısına göre en iyi tahminleme sonucunu, Lojistik Regresyon ile kurulan model vermiştir. Derse katılma yöntemlerinin tümünün (Senkron, Asenkron ve Hibrit) öğrenci başarısına etkisi Karışıklık Matrisi yöntemiyle karşılaştırılmıştır ve en güvenilir yöntemin Hibrit olduğu görülmüştür. Bu çalışma ile üniversitelerde derse katılma yöntemlerinden hangisinin, öğrenciler açısından daha güvenilir olacağına yönelik çıkarımlarda bulunulmuştur. Dolayısıyla yapılan çıkarımlar ile birlikte bir dahaki akademik dönem için öğrenci başarı düzeyinin artması yönünde, derse katılma yöntemlerinden hangisinin daha güvenilir olduğu konusunda tahminleme mümkün olmuştur.
- 26 January 2024
- Published by Duzce Universitesi Bilim ve Teknoloji Dergisi in Düzce Üniversitesi Bilim ve Teknoloji Dergisi
- Vol. 12 (1) , 89-111
Open AccessPandemi Sürecinde Uzaktan Eğitimde Senkron, Asenkron Ve Hibrit Yapılmış Derslerde Veri Madenciliği İle Öğrenci Performans Analizi
Günümüzde üniversite öğrencilerinin eğitime ve eğitim materyallerine internetten erişim oranları oldukça artmıştır. Eğitimde internetin kullanılması ve ders materyallerine erişimin artmasına bağlı olarak analiz edilebilecek veri setinde artış meydana gelmiştir. Bu veri setlerinden bir tanesi de (planlı veya acilen) uzaktan eğitim sürecine geçen üniversitelerin, uzaktan eğitim sistemlerinde biriken ödev, sınav, proje, performans, devam notları ve benzeridir. Yeni Korona Virüs (Covid-19) pandemisinde Yüksek Öğretim Kurumu’nun (YÖK) tavsiyesi ile üniversiteler eğitimlerine uzaktan Asenkron, Senkron ve Hibrit yöntemlerini kullanarak devam etmiş, hatta sınavları uzaktan eğitim sisteminde yapmak zorunda kalmışlardır. Bu araştırmada, Kayseri Üniversitesinin Uzaktan Eğitim Uygulama ve Araştırma Merkezi (KAYUZEM) sisteminden alınmış veriler kullanılmıştır. Araştırma kapsamında 8319 işlenmiş veri bulunmaktadır. Bu veriler üzerinde Veri Madenciliği (VM) alanında kullanılan RapidMiner programının otomatik modelleme özelliği kullanılarak varsayılan algoritmalarla geleceğe yönelik tahminleme işlemi yapılmıştır. Varsayılan algoritmalar arasından en iyi sonucu veren Derin Öğrenme, Naive Bayes, Gradient Boosted Trees, Lojistik Regresyon kullanılmış ve bu otomatik modelleme de yer almayan k-En Yakın Komşu (k-NN) algoritması da çalışmaya dahil edilmiştir. Bu 5 algoritmanın parametreleri üzerinde değişiklikler yapılarak daha iyi sonuçlar elde edilmeye çalışılmıştır. Öğrenci başarısına göre en iyi tahminleme sonucunu, Lojistik Regresyon ile kurulan model vermiştir. Derse katılma yöntemlerinin tümünün (Senkron, Asenkron ve Hibrit) öğrenci başarısına etkisi Karışıklık Matrisi yöntemiyle karşılaştırılmıştır ve en güvenilir yöntemin Hibrit olduğu görülmüştür. Bu çalışma ile üniversitelerde derse katılma yöntemlerinden hangisinin, öğrenciler açısından daha güvenilir olacağına yönelik çıkarımlarda bulunulmuştur. Dolayısıyla yapılan çıkarımlar ile birlikte bir dahaki akademik dönem için öğrenci başarı düzeyinin artması yönünde, derse katılma yöntemlerinden hangisinin daha güvenilir olduğu konusunda tahminleme mümkün olmuştur.