Botnet attack detection with machine learning methods: Dimension reduction approaches

In this study, an attack detection system was developed to detect botnet attacks, a significant threat in cybersecurity. The study, conducted on the CSE-CIC-IDS2018 dataset, utilized machine learning algorithms such as k-nearest neighbors, logistic regression, decision tree, and random forest models to create an anomaly-based classification method. Principal component analysis, correlation-based, and mutual information dimensionality reduction methods were applied to process the high-dimensional features in the dataset, and hybrid methods combining the strengths of these techniques were developed. By using dimensionality reduction methods, unnecessary features in the dataset were eliminated, optimizing the classification time and improving model performance. Hybrid methods, combining the advantages of principal component analysis, correlation-based, and mutual information techniques, reduced both the processing time and provided high performance in terms of accuracy and F1 score. This approach effectively selects important features in complex data structures while shortening the model training time, making it suitable for real-time attack detection systems. The findings of the study show that hybrid methods significantly reduced processing time while maintaining high performance compared to other dimensionality reduction techniques. The results offered a comprehensive evaluation in terms of accuracy, precision, sensitivity, F1 score, and training time. In particular, hybrid methods proved to be an effective tool in cybersecurity by balancing speed and accuracy. This study developed an approach aimed at reducing classification time while maintaining accuracy and demonstrated that hybrid methods are an effective solution to improve the performance of attack detection systems. The study contributes significantly to the development of fast, efficient, and high-accuracy attack detection systems in cybersecurity.