Olcay Alpay

Olcay Alpay

Auto-generated profile, based on publicly available information

Olcay Alpay

Sinop University

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
30 September 2022 in Academic Platform Journal of Engineering and Smart Systems
Verified
Journal Article
Closed access
PDF file
OAOlcay Alpay
Olcay Alpay
NDNazan Danacioğlu
Nazan Danacioğlu
EÇEmel Çankaya
Emel Çankaya
Citation is considered as the most popular quality assessment metric for scientific papers, and it is thus important to determine what factors promote the citation count of a paper in comparison to the others in the same field. The main aim of this study is to model the citation counts of the research published in SCI or SCI-Expanded journals of Statistics field with the growing number of scientific works in Turkey. It is well known that the right-skewed nature of the counts makes the classical regression modelling inappropriate, even if the log transformation of counts is applied [1]. Due to the fact that distribution of citation counts involves a great number of zeros, this study serves for an additional aim that is to model the counts with advanced discrete regression models for a more precise prediction [2]. Data collected for this study consist of the citation counts of all scientific papers produced by 261 Statisticians in between 2005-2017. Discrete models varying from Poisson to Zero-Inflated or Hurdle were constructed by possible influential factors, such as the publication age, the number of references, the journal category etc. Predictive performances of alternative discrete models were compared via AIC and Vuong test [3]. Results suggested that Zero Inflated Negative Binomial and Hurdle Negative Binomial mixture models are the best forms to predict the zero inflation of citation counts [4]. In addition, the influential factors of the final model were interpreted to make some suggestions to Statisticians to increase the citation counts of their work.

29 June 2024 in Sinop Üniversitesi Fen Bilimleri Dergisi
Verified
Journal Article
Open access
OAOlcay Alpay
Olcay Alpay
Bu çalışma, iki durumlu olayları modellemek için kullanılan makine öğrenmesi tekniklerinde karşılaşılan nadirlik ve “çoklu doğrusal bağlantı” ya da sadece “çoklu bağlantı” olarak tanımlanan sorunu ele alınmaktadır. Çoklu doğrusal bağlantı (ÇDB), bağımsız değişkenler arasında bir ya da birden fazla kuvvetli doğrusal bağımlılık olma durumudur ve bir sorun olarak ortaya çıkar. Üzerinde çalışılan veri içerisinde çoklu doğrusal bağlantı probleminin var olması regresyon katsayılarının varyanslarının büyümesi gibi olumsuz bir sonuca sebebiyet verir. Bu çalışmada, Lasso ve Ridge Regresyon ile GradientBoost, XGBoost, LightGBM ve AdaBoost gibi artırma algoritmaları içeren düzenleme ve ölçeklendirme tekniklerinin, çoklu doğrusal bağlantılı nadir olayların modellenmesinde, algoritmaların performanslarını karşılaştırmak için detaylı bir simülasyon çalışması sunulmaktadır. Simülasyon çalışmasında, verideki dengesizliği ortadan kaldırmak amacıyla yeniden örnekleme yöntemleri kullanılarak sonuçlara etkisi Hata Kareler Ortalaması (HKO), R^2, Hassasiyet (Precision-Prec), Duyarlılık (Recall-Rec) ve Eğri Altında Kalan Alan (Area Under the Curve-AUC) gibi performans metrikleri ve İşlem Karakteristik Eğrisi (Receiver Operating Characteristic- ROC) grafikleri ile araştırılmaktadır. Sonuçlar Lasso, Ridge ve Boosting algoritmalarının ÇDB’ya sahip nadir olayların modellenmesinde hangi yöntemin uygun olduğunu belirlemek açısından katkı sunmaktadır.

30 December 2025 in Cumhuriyet Science Journal
Verified
Journal Article
Closed access
OAOlcay Alpay
Olcay Alpay
As data science and machine learning continue to evolve, binary event classification has become increasingly important. Logistic Regression (LR) is a standard baseline, yet it can underestimate probabilities in rare-event settings. This study combines a 16-scenario simulation (rarity 5–10%, n∈ {1000,5000}, p∈ {3,5,7,10}, 100 repeats) with a real-world application to assess Support Vector Machines (SVM), Random Forest (RF), and Gradient Boosting (GB) as alternatives to LR. Training data were balanced using the SMOTETomek hybrid method. In simulation, LR attained the highest balanced performance (G_mean) across cases, with GB the closest competitor; SVM lagged, and RF yielded the lowest G_mean despite often leading test accuracy and precision. On a wine dataset adjusted to 5% and 10% rarity, RF/GB achieved top test accuracy/recall (e.g., ACC=0.998/0.997 with REC=0.958/0.989 at 5% and ACC=0.998/0.994 with REC=0.989/0.977 at 10%), mirroring their strong aggregate accuracy. Overall, the “best” model depends on the target metric: LR/GB when balanced minority–majority performance is critical, and RF when overall accuracy/precision is prioritized.

1 June 2026 in Hacettepe Journal of Mathematics and Statistics
Verified
Journal Article
Closed access
OAOlcay Alpay
Olcay Alpay
While numerous oversampling strategies have emerged in recent years, synthesizing minority-class samples that preserve the intrinsic data structure while improving classifier performance remains challenging, especially under severe class imbalance. In this study, we propose HypoGAN, a generative oversampling framework for imbalanced numeric tabular data that integrates minority mode discovery, local minority-majority boundary information, latent-noise-driven candidate generation, and post-generation filtering to produce classification-useful minority samples. To evaluate the proposed framework, we adopted a two-phase experimental strategy: (i) simulation experiments under class imbalance ratios of 90:10, 95:5, and 98:2 with feature dimensionalities of n = 3, 5, 10, 20, and (ii) real-world experiments on Wisconsin Breast Cancer, Pima Indians Diabetes, and detection of credit card fraud. In all experiments, HypoGAN was compared with SMOTE, ADASYN, and Borderline-SMOTE within a nested train/validation/test framework with leakage protection. The results indicate that HypoGAN is a competitive oversampling framework in a range of challenging imbalance settings. In the simulation study, it achieved particularly strong performance in lower- and medium-dimensional scenarios, while remaining precision-oriented under more difficult conditions. In real-world experiments, HypoGAN remained competitive in the Wisconsin Breast Cancer and Credit Card Fraud Detection datasets, achieving F1-scores of 0.9489 and 0.9028, respectively, compared to 0.9489 and 0.9037 for SMOTE. Additional results suggest that HypoGAN’s performance is scenariodependent, with effectiveness influenced by the dataset’s structure and the characteristics of the imbalance.