@article{article_1698023, title={Network Traffic Factors Detection and Prediction Using Artificial Neural Networks}, journal={Sakarya University Journal of Computer and Information Sciences}, volume={8}, pages={688–700}, year={2025}, DOI={10.35377/saucis...1698023}, url={https://izlik.org/JA75BH97FT}, author={Özdemir, Muhammed and Akpınar, Mustafa and Eski, Hüseyin}, keywords={Cyber security, Firewall, Log, Linear regression, Principal component analysis, Artificial neural networks}, abstract={Rising technology generates an increasing data flow on the internet by the day. Specific security systems are crucial for managing this growing data traffic. One such security system is the firewall used in Turkey Maritime Enterprises Inc. (TME). All TME internet traffic is controlled by a security firewall that operates within existing rules and regulations. The security firewall produces output based on traffic categorized as “allowed” or “blocked.” This study collected log records from the firewall on five different days to create datasets. Twenty-six variables were extracted from the firewall logs, excluding the output. The study aimed to identify the most significant parameters impacting the output using linear regression (LR) and principal component analysis (PCA). The analysis identified mutually influential log variables affecting the output using two methods. After, the initial dataset with 26 variables and a reduced dataset with six variables were used to predict output using the artificial neural network (ANN) for five datasets. The prediction accuracy and precision ranges were between 85% and 88% and 92% and 98%, respectively. The F1-score showed results between 89% and 92% in addition to accuracy and precision. ML and PCA methods successfully identified crucial variables for estimating output and reduced the number of variables from 26 to 5. Moreover, it was noted that the ANN could accurately determine whether firewall traffic would be blocked or allowed based on the reduced datasets. Reducing the feature space from 26 to 6 variables identified via MLR+PCA improved ANN performance across hours, indicating that compact, interpretable inputs can support accurate firewall-traffic prediction.}, number={4}