TITLE:
Cybersecurity Modeling: Application of Optimization in Machine Learning-Based Detection Systems
AUTHORS:
Gloria A. Odiaga, Newton Masinde, Castro Yoga
KEYWORDS:
Optimization, Cybersecurity, Machine Learning, Deep Learning, Constraint, Domain Set, Objective Function, Variable, Cyberattack
JOURNAL NAME:
Journal of Computer and Communications,
Vol.13 No.9,
September
28,
2025
ABSTRACT: Mathematical optimization is a fundamental aspect of machine learning (ML). An ML task can be conceptualized as optimizing a specific objective using the training dataset to discern patterns and to predict and generalize on new, unseen data through testing. The reliability of optimization in addressing these tasks is fundamental to the efficacy of ML solutions, which has prompted the exploration of complex ML tasks with advanced mathematical approaches and substantially larger datasets. As dataset sizes grow, training complex models can take longer, even with high-performance hardware, qualifying the need for efficient optimization techniques and the performance enhancement of existing optimization methods. This study discusses optimization techniques and their application in ML and deep learning (DL). By framing the detection task as an optimization problem, the study proposes a systematic framework that includes mathematical modeling of the problem. The study also emphasizes the importance of selecting appropriate optimization methods across model development in architectural design, training, and tuning procedures, grounded in the mathematical modeling of a cyberattack detection task to ensure optimal performance in securing web-based systems. The experimental results on a Long Short-Term Memory (LSTM) model show an accuracy of 0.947, a False Negative (FN) rate of 0.053, and a False Positive (FP) rate of 0.011, hence demonstrating that integrating the proposed framework in cybersecurity model design can enhance the attack detection performance.