Browsing by Author "Sever, H."

Now showing 1 - 4 of 4

Citation - Scopus: 0
A Data Fusion Approach In Protein Homology Detection
(2008) Oğul, Hasan; Polatkan, A.C.; Oǧul, H.; Sever, Hayri; Sever, H.; 11916; Bilgisayar Mühendisliği
The discriminative framework for protein remote homology detection based on support vector machines (SVMs) is reconstructed by the fusion of sequence based features. In this respect, n-peptide compositions are partitioned and fed into separate SVMs. The SVM outputs are evaluated with different techniques and tested to discern their ability for SCOP protein super family classification on a common benchmarking set. It reveals that the fusion approach leads to an improvement in prediction accuracy with a remarkable gain on computer memory usage. © 2008 IEEE.
Citation - Scopus: 2
Application of Artificial Intelligence in Early–Stage Diagnosis of Sepsis
(Association for Computing Machinery, 2022) Par, O.E.; Sever, Hayri; Sezer, E.A.; Sever, H.; 11916; Bilgisayar Mühendisliği
Patient care is a critical task, which requires a lot of effort. Medical practitioners face many challenges, especially during diagnosing different diseases. Sepsis is one of the riskiest diseases, which proves to be lethal for Intensive Care Unit (ICU) patients. World Health Organization (WHO) has declared it a major cause of death worldwide. Early-stage diagnosis of sepsis can help in terminating it in the start. But unfortunately, medical practitioners encounter hitches in the early-stage diagnosis of sepsis. The study used SOFA (Sequential Organ Failure Assessment) for measuring the severity of sepsis in patients. The study employs artificial intelligence techniques such as Multilayer Perceptron (MLP) and Random Forest (RF) to diagnose early-stage of sepsis. The study compared the performance of MLP (connected and non-connected) and Random Forest (connected and non-connected) algorithms. The results indicate that for both of the algorithms, the connected method yielded better results than the non-connected method. Further, it was found that RF both connected and non-connected algorithms yielded better results than MLP algorithms and the Random Forest connected algorithm yielded highly accurate results for diagnosing early-stage sepsis in the 3rd hour. © 2022 ACM.
Citation - WoS: 0
Citation - Scopus: 0
Production and Retrieval of Rough Classes in Multi Relations
(Ieee Computer Soc, 2007) Tolun, M.R.; Sever, H.; Gorur, A.K.
Organizational memory in today's business world forms basis for organizational learning, which is the ability of an organization to gain insight and understanding from experience through experimentation, observation, analysis, and a willingness to examine both successes and failures. This basically requires consideration of different aspects of knowledge that may reside on top of a conventional information management system. Of them, representation, retrieval and production issues of meta patterns constitute to the main theme of this article. Particularly we are interested in a formal approach to handle rough concepts. We utilize rough classifiers to propose a preliminary framework based on minimal term sets with p-norms to extract meta patterns. We describe a relational rule induction approach, which is called rila. Experimental results are provided on the mutagenesis, and the KDD Cup 2001 genes data sets. © 2007 IEEE.
Citation - Scopus: 1
Text-Based Fake News Detection via Machine Learning
(Springer Science and Business Media Deutschland GmbH, 2021) Mertoğlu, U.; Sever, Hayri; Genç, B.; Sever, H.; 11916; Bilgisayar Mühendisliği
The nature of information literacy is changing as people incline more towards using digital media to consume content. Consequently, this easier way of consuming information has sparked off a challenge called “Fake News”. One of the risky effects of this notorious term is to influence people’s views of the world as in the recent example of coronavirus misinformation that is flooding the internet. Nowadays, it seems the world needs “information hygiene” more than anything. Yet real-world solutions in practice are not qualified to determine verifiability of the information circulating. Presenting an automated solution, our work provides an adaptable solution to detect fake news in practice. Our approach proposes a set of carefully selected features combined with word-embeddings to predict fake or valid texts. We evaluated our proposed model in terms of efficacy through intensive experimentation. Additionally, we present an analysis linked with linguistic features for detecting fake and valid news content. An overview of text-based fake news detection guidance derived from experiments including promising results of our work is also presented in this work. © 2021, The Author(s), under exclusive license to Springer Nature Switzerland AG.