COMPARATIVE ANALYSIS OF MACHINE LEARNING ALGORITHMS FOR PREDICTING CYBERSECURITY ATTACK SUCCESS: A PERFORMANCE EVALUATION

Аннотация

INTRODUCTION are employed to analyze historical attack data, The rapid advancement of technology has identify patterns, and predict future threats dramatically transformed various aspects of (Sommer & Paxson, 2019). The application of ML modern life, introducing both opportunities and techniques has been shown to enhance threat challenges. One of the most pressing concerns in detection, streamline incident response, and the digital age is cybersecurity, as the proliferation improve overall security posture (Li et al., 2020). of interconnected systems and data has increased Among the various ML algorithms available, vulnerability to cyberattacks. These attacks, Logistic Regression (LR), Random Forest (RF), ranging from data breaches to sophisticated Support Vector Machine (SVM), Gradient Boosting ransomware campaigns, pose significant risks to (GB), and K-Nearest Neighbors (KNN) are organizations and individuals, underscoring the frequently used in cybersecurity research due to need for effective defense mechanisms (NIST, their distinct advantages and capabilities. 2021). In response to this growing threat, researchers and practitioners have increasingly Logistic Regression, a fundamental classification turned to machine learning (ML) as a powerful tool algorithm, models the probability of a binary for predicting and mitigating cybersecurity risks. outcome based on predictor variables (Menard, Machine learning, a subset of artificial intelligence 2021). It is valued for its simplicity and (AI), involves the development of algorithms that interpretability, making it a popular choice for can learn from and make predictions based on initial analyses of attack success. Random Forest, data. In the context of cybersecurity, ML models an ensemble learning method that constructs https://www.theamericanjournals.com/index.php/tajet 82 THE USA JOURNALS THE AMERICAN JOURNAL OF ENGINEERING AND TECHNOLOGY (ISSN – 2689-0984) VOLUME 06 ISSUE09 multiple decision trees and aggregates their LITERATURE REVIEW results, excels in handling large datasets and The application of machine learning in capturing complex, non-linear relationships (Liaw cybersecurity has been extensively studied, & Wiener, 2002). Its robustness and accuracy have highlighting its potential in threat detection and made it a favored choice for various cybersecurity prediction. Machine learning algorithms have been applications. Support Vector Machines, known for utilized to analyze various types of attack data, their effectiveness in high-dimensional spaces, are such as network traffic, system logs, and designed to find the optimal hyperplane that vulnerability assessments, to identify potential separates different classes (Cortes & Vapnik, threats and improve response strategies. 1995). SVMs are particularly useful for Logistic Regression (LR), a widely used statistical classification tasks involving intricate patterns, method, has shown utility in binary classification although they require careful parameter tuning tasks, including cybersecurity threat prediction. It and data scaling. Gradient Boosting, another models the probability of a binary outcome based ensemble technique, builds models sequentially to on one or more predictor variables. Studies have correct errors made by previous models, demonstrated its effectiveness in predicting attack enhancing predictive accuracy and performance success by analyzing features such as attack type (Friedman, 2001). This method is effective in and network traffic (Gao et al., 2018). addressing imbalanced data and capturing subtle patterns. Random Forest (RF), an ensemble learning method, aggregates multiple decision trees to K-Nearest Neighbors, a straightforward algorithm improve prediction accuracy and robustness. It has that classifies instances based on their proximity to been frequently applied in cybersecurity due to its neighboring data points, is valued for its simplicity ability to handle large datasets and capture and ease of implementation (Cover & Hart, 1967). complex, non-linear relationships. Research However, KNN may struggle with high- indicates that RF can effectively classify and dimensional and imbalanced datasets, which are predict various types of cyberattacks with high common in cybersecurity scenarios. Despite the accuracy and robustness (Kang et al., 2019). potential benefits of these algorithms, their effectiveness can vary depending on the nature of Support Vector Machines (SVM), known for their the dataset, the specific characteristics of the classification capabilities, have been used in attacks, and the computational resources available. cybersecurity to distinguish between malicious Consequently, a comparative analysis of these ML and benign activities. SVMs are effective in models is crucial for identifying the most effective handling high-dimensional data and finding approach for predicting cybersecurity attack optimal decision boundaries. However, their success. This study aims to