WoS İndeksli Yayınlar Koleksiyonu
Permanent URI for this collectionhttps://hdl.handle.net/20.500.12573/394
Browse
8 results
Search Results
Conference Object Ensemble Churn Prediction for Internet Service Provider with Machine Learning Techniques(IEEE, 345 E 47TH ST, NEW YORK, NY 10017 USA, 2020) Goy, Gokhan; Kolukisa, Burak; Bahcevan, Cenk; Gungor, Vehbi CagriWith the developing technology in every fields, a competitive marketing environment has been arised In this competitive environment analyzing customer behavior has become vital In particular, the ability to easily change any service provider has become vet) , critical for the company to continue its existence At the same time, the amount of financial resources spent on retaining instituters much less than to obtain new clients. In this context, the traditional methods of examining vast amount of data obtained today for establishing decision support systems have lost their validities In this study. we used a dataset which is provided by TurkNet serving as an internet service provider in Turkey. Various preprocessing steps has performed on this dataset and then classification algorithms ran. Afterwards results have obtained and compared. The results of these experiments analyzed in terms of the area under the curve value In this context the aunt successful classifier algorithm has been determined as the Random Trees algorithm with a value of 0.936.Article Citation - WoS: 17Citation - Scopus: 21Machine Learning Approaches for Underwater Sensor Network Parameter Prediction(Elsevier, 2023-05) Uyan, Osman Gokhan; Akbas, Ayhan; Gungor, Vehbi CagriUnderwater Acoustic Sensor Networks (UASNs) have recently attracted scientists due to its wide range of real -world applications. However, there are design challenges in UASNs, such as limited network lifetime and low communication reliability provoked by the constrained battery supply of sensors and harsh channel conditions in the underwater environments. To meet communication reliability requirements, packet-duplication and multi -path routing algorithms have been recommended in the literature. Furthermore, underwater sensors may convey sensitive data, which must be masked to avoid eavesdropping attempts. To improve network security, cryptographic encryption is the most widely used method. Nevertheless, data encryption needs computations to cipher the data, which consumes extra energy, resulting in a cutback in the life span of the network. To address these challenges, an optimization model has been proposed to evaluate the impacts of multi-path routing, packet duplication, encryption, and data fragmentation on the lifetime of the UASNs. However, the solution time of the proposed optimization model is quite high, and sometimes it cannot come up with feasible solutions. To this end, in this study, different regression and neural network methods have been proposed to predict network param-eters and energy consumptions of underwater nodes as supplementary methods to optimization models. Per-formance evaluations show that the proposed methods yield remarkably accurate predictions and can be used for energy consumption prediction in UASNs.Article Citation - WoS: 10Citation - Scopus: 8Ensemble Feature Selection for Clustering Damage Modes in Carbon Fiber-Reinforced Polymer Sandwich Composites Using Acoustic Emission(Wiley-VCH Verlag GmbH, 2024-07-15) Gulsen, Abdulkadir; Kolukisa, Burak; Caliskan, Umut; Bakir-Gungor, Burcu; Gungor, Vehbi CagriAcoustic emission (AE) serves as a noninvasive technique for real-time structural health monitoring, capturing the stress waves produced by the formation and growth of cracks within a material. This study presents a novel ensemble feature selection methodology to rank features highly relevant with damage modes in AE signals gathered from edgewise compression tests on honeycomb-core carbon fiber-reinforced polymer. Two distinct features, amplitude and peak frequency, are selected for labeling the AE signals. An ensemble-supervised feature selection method ranks feature importance according to these labels. Using the ranking list, unsupervised clustering models are then applied to identify damage modes. The comparative results reveal a robust correlation between the damage modes and the features of counts and energy when amplitude is selected. Similarly, when peak frequency is chosen, a significant association is observed between the damage modes and the features of partial powers 1 and 2. These findings demonstrate that, in addition to the commonly used features, other features, such as partial powers, exhibit a correlation with damage modes. This article presents a novel ensemble feature selection methodology to rank features relevant to damage modes on acoustic emission signals in carbon fiber-reinforced polymer sandwich composites. Subsequently, ranked features are utilized in unsupervised clustering models to identify damage modes. The comparative results demonstrate that, along with common features, other features, like partial powers, have a robust correlation with damage modes.image (c) 2024 WILEY-VCH GmbHConference Object Citation - Scopus: 4Deep Learning Based Employee Attrition Prediction(Springer International Publishing AG, 2023) Gurler, Kerem; Pak, Burcu Kuleli; Gungor, Vehbi CagriEmployee attrition is a critical issue for the business sectors as leaving employees cause various types of difficulties for the company. Some studies exist on examining the reasons for this phenomenon and predicting it with Machine Learning algorithms. In this paper, the causes for employee attrition is explored in three datasets, one of them being our own novel dataset and others obtained from Kaggle. Employee attrition was predicted with multiple Machine Learning and Deep Learning algorithms with feature selection and hyperparameter optimization and their performances are evaluated with multiple metrics. Deep Learning methods showed superior performances in all of the datasets we explored. SMOTE Tomek Links were utilized to oversample minority classes and effectively tackle the problem of class imbalance. Best performing methods were Deep Random Forest on HR Dataset from Kaggle and Neural Network for IBM and Adesso datasets with F1 scores of 0.972, 0.642 and 0.853, respectively.Conference Object Citation - WoS: 2Credit Card Fraud Detection With Machine Learning Methods(IEEE, 2019-09) Goy, Gokhan; Gezer, Cengiz; Gungor, Vehbi CagriWith the increase in credit card usage of people, the credit card transactions increase dramatically. It is difficult to identify fraudulent transactions among the vast amount of credit card transactions. Although credit card fraud is limited in number of transactions, it causes serious problems in terms of financial losses for individuals and organizations. Even though large number of studies has been conducted to solve this problem, there is no generally accepted solution. In this paper, a publicly available data set is used. The unbalance problem of the data set was solved by using hybrid sampling methods together. On this data set, comparative performance evaluations have been conducted. Different from other studies, the Area Under the Curve (AUC) metric, which expresses the success in such data sets, has also been used in addition to standard performance metrics. Since it is also important to quickly detect credit card fraud transactions; the running time of different methods is also presented as another performance metric.Article Citation - WoS: 25Citation - Scopus: 41An Efficient Network Intrusion Detection Approach Based on Logistic Regression Model and Parallel Artificial Bee Colony Algorithm(Elsevier, 2024-04) Kolukisa, Burak; Dedeturk, Bilge Kagan; Hacilar, Hilal; Gungor, Vehbi CagriIn recent years, the widespread use of the Internet has created many issues, especially in the area of cybersecurity. It is critical to detect intrusions in network traffic, and researchers have developed network intrusion and anomaly detection systems to cope with high numbers of attacks and attack variations. In particular, machine learning and meta-heuristic methods have been widely used for network intrusion detection systems (NIDS). However, existing studies on these systems usually suffer from low performance results such as accuracy, F1-measure, false positive rate, and false negative rate, and generally do not use automatic parameter tuning techniques. To address these challenges, this study proposes a novel approach based on a logistic regression model trained using a parallel artificial bee colony (LR-ABC) algorithm with a hyper-parameter optimization technique. The performance of the proposed model is evaluated against state -of-the-art machine learning and deep learning models on two publicly available NIDS datasets. Comparative performance evaluations show that the proposed method achieved satisfactory results with accuracy of 88.25% on the UNSW-NB15 dataset and 90.11% on the NSL-KDD dataset, and F1-measures of 88.26% and 90.15%, respectively. These findings demonstrate the efficacy of the proposed LR-ABC model in enhancing the accuracy and reliability, while providing a scalable solution to adapt to the dynamic and evolving landscape of cybersecurity threats.Article Citation - WoS: 23Citation - Scopus: 35A Review of On-Device Machine Learning for IoT: An Energy Perspective(Elsevier, 2024-02) Tekin, Nazli; Aris, Ahmet; Acar, Abbas; Uluagac, Selcuk; Gungor, Vehbi CagriRecently, there has been a substantial interest in on-device Machine Learning (ML) models to provide intelligence for the Internet of Things (IoT) applications such as image classification, human activity recognition, and anomaly detection. Traditionally, ML models are deployed in the cloud or centralized servers to take advantage of their abundant computational resources. However, sharing data with the cloud and third parties degrades privacy and may cause propagation delay in the network due to a large amount of transmitted data impacting the performance of real-time applications. To this end, deploying ML models on-device (i.e., on IoT devices), in which data does not need to be transmitted, becomes imperative. However, deploying and running ML models on already resource-constrained IoT devices is challenging and requires intense energy consumption. Numerous works have been proposed in the literature to address this issue. Although there are considerable works that discuss energy-aware ML approaches for on-device implementation, there remains a gap in the literature on a comprehensive review of this subject. In this paper, we provide a review of existing studies focusing on-device ML models for IoT applications in terms of energy consumption. One of the key contributions of this study is to introduce a taxonomy to define approaches for employing energy-aware on-device ML models on IoT devices in the literature. Based on our review in this paper, our key findings are provided and the open issues that can be investigated further by other researchers are discussed. We believe that this study will be a reference for practitioners and researchers who want to employ energy-aware on-device ML models for IoT applications.Conference Object Makine Öğrenmesi Teknikleri ile İnternet Servis Sağlayicisi için Müşteri Kayip Tahmini(IEEE, 2020) Goy, Gokhan; Kolukisa, Burak; Bahcevan, Cenk; Gungor, Vehbi CagriWith the developing technology in every fields, a competitive marketing environment has been arised In this competitive environment analyzing customer behavior has become vital In particular, the ability to easily change any service provider has become vet) , critical for the company to continue its existence At the same time, the amount of financial resources spent on retaining instituters much less than to obtain new clients. In this context, the traditional methods of examining vast amount of data obtained today for establishing decision support systems have lost their validities In this study. we used a dataset which is provided by TurkNet serving as an internet service provider in Turkey. Various preprocessing steps has performed on this dataset and then classification algorithms ran. Afterwards results have obtained and compared. The results of these experiments analyzed in terms of the area under the curve value In this context the aunt successful classifier algorithm has been determined as the Random Trees algorithm with a value of 0.936.