Yazar "Kilic, Fatih" seçeneğine göre listele

Listeleniyor 1 - 3 / 3
  • [ X ]
    Öğe
    Discrete Starfish Optimization Algorithm for Symmetric Travelling Salesman Problem
    (IEEE-Inst Electrical Electronics Engineers Inc, 2025) Aktas, Muhammet; Kilic, Fatih
    This paper introduces a new discrete StarFish Optimization Algorithm (D-SFOA) to solve a complex discrete Symmetric Travelling Salesman Problem (STSP). The discrete SFOA algorithm is initialized with the initial population of classical SFOA, and the continuous values of the individuals in the population are converted to the discrete version using the random key method. Ten neighbourhood methods used in this study provide diversity to the starfish population, and the 2-opt local search algorithm allows the study to find shorter tours. The performance of D-SFOA is tested on STSP samples ranging in size from 30 to 1084 from TSPLIB. Discrete versions of the Grey Wolf Optimizer (D-GWO) and Harris Hawk Optimization (D-HHO) algorithms are applied with the same parameters to compare the performance of the proposed algorithm. The algorithm uses descriptive statistics such as average tour, best tour, percentage of deviation of the mean tour, percentage of deviation of the best tour, and execution time to ensure a fair comparison. The Wilcoxon signed-rank test and Ablation test are applied to measure the significant difference in the values of the algorithms and to observe the performance effect of the main components used in the proposed algorithm on tour length and execution time, respectively. This study's numerical and statistical results show that D-SFOA has significantly outperformed other alternative algorithms and provided better solutions than the best-known solution.
  • [ X ]
    Öğe
    Enhanced Phishing Website Detection Through Effective Feature Selection With Time-Varying Mirrored S-Shaped Transfer Function
    (Wiley, 2025) Kilic, Fatih; Erdi, Kemal; Aktas, Muhammet
    The increasing use of websites and the increasing sophistication of cybercrimes have highlighted the need for effective prevention mechanisms. Phishing attacks represent a significant cybersecurity threat, as they exploit human weaknesses and deceptive web-based tactics to achieve unauthorized access to sensitive user information. Due to the evolving and increasingly sophisticated methodologies cybercriminals use, there is an urgent demand for advanced and innovative approaches to detect and mitigate these threats accurately. In this paper, a novel approach is proposed by incorporating a time-varying mirrored S-shaped transfer function into three prominent binary optimization algorithms: Binary Grey Wolf Optimization (TVBGWO), Binary Particle Swarm Optimization (TVBPSO), and Binary Harris Hawk Optimization (TVBHHO). The proposed models are rigorously evaluated using a publicly available phishing website dataset. Using a time-varying mirrored S-shaped transfer function for feature selection enables metaheuristic algorithms to improve the balance between exploration and exploitation, thus enhancing their ability to classify phishing websites accurately. The Friedman test confirms statistically significant differences among the compared methods (X-2 = 70.3, p = 8.87 x 10(-14)). The Wilcoxon Rank-Sum test is applied to determine the difference in the performance of the time-varying methods among the standard BGWO, BHHO, and BPSO algorithms in this study. TV-based variants of BPSO and BGWO showed a solid and balanced performance, achieving the highest test accuracy (90.3% and 90.2%) and F1-scores (0.91%). Proposed models achieve promising high accuracy and F1-score performance in detecting phishing websites using fewer features and suggesting their potential.
  • [ X ]
    Öğe
    Experimental Investigation of Artificial Intelligence Models for Recommender Systems
    (IEEE, 2025) Ozbey, Onur; Kilic, Fatih
    The increasing amount of information and content on the Internet has made it more difficult for users to access items that may interest them. Recommender systems have emerged to determine items that may affect users by filtering information obtained from users' behaviors through certain filters and presenting them to the user. This study presents the performance of widely preferred recommendation system models such as artificial neural network, XGBoost, KNNRegressor, and LightGBM models in the related literature. The well-known datasets, namely MovieLens-100k and MovieLens-32M, are used to evaluate these models. According to the presented results, it was observed that the artificial neural network model was ahead of other models in terms of performance data.