Published Paper Details:

Network anomaly detection is an essential element of contemporary cybersecurity infrastructure, especially as organizations encounter more and more sophisticated and changing cyber threats. Conventional intrusion detection systems, which rely intensely on static signature databases and preconfigured rule sets, exhibit clear shortcomings in discovering new attack patterns and responding to dynamic threat environments. This research explores the optimization and utilization of conventional machine learning models for improved network anomaly detection, filling the research gap in which deep learning methods have dominated the potential that classical ML strategies can offer. With in-depth scrutiny of the UNSW-NB15 dataset comprising contemporary attack patterns in nine unique categories, this study compares six supervised machine learning classifiers: Logistic Regression, Decision Trees, Random Forests, Gaussian Naive Bayes, K-Nearest Neighbors (KNN), and LightGBM. The approach utilizes stringent data preprocessing pipelines, systematic feature engineering, and strenuous performance analysis across diverse metrics such as accuracy, precision, recall, and F1-score. Experimental findings show outstanding performance attainment, as Decision Trees, Random Forests, and LightGBM each recorded perfect accuracy at 100% and precision at 1.000, while KNN recorded 99.68% accuracy with 0.9968 precision and Gaussian Naive Bayes recorded 97.54% accuracy with 0.9764 precision. These results undermine the common belief that deep learning architectures of high complexity are required for successful anomaly detection, as it is shown that conventional ML models, if they are well optimized, can perform better with benefits in computational efficiency, interpretability, and deployment practicability. The work provides substantial practical implications for cybersecurity applications, especially in resource-limited settings and edge computing environments where light, interpretable models are desired. This research lays the groundwork for future-generation intrusion detection systems that strike a balance between outstanding detection rates and operational efficiency, offering actionable intelligence for security professionals while propelling the science of intelligent network security through evidence-based optimization of standard machine learning techniques.