Machine Learning-Driven Intrusion Detection Systems: Reducing False Alarms and Enhancing Accuracy

Abstract

The increasing sophistication of cyber threats presents ongoing challenges for securing modern networks, particularly in addressing the limitations of Intrusion Detection Systems (IDS). Traditional IDS solutions often suffer from high false-positive rates and limited accuracy in detecting novel or unknown attacks, leading to inefficiencies in security management. This paper explores the use of multiple Machine Learning (ML) algorithms to improve IDS performance, focusing on models such as Artificial Neural Networks (ANN), K-Nearest Neighbors (KNN), Decision Trees (DT), Naive Bayes (NB), Logistic Regression (LR), and Support Vector Machines (SVM). The research employs the KDD Cup 1999 dataset, a well-known benchmark for intrusion detection, to evaluate the effectiveness of these models. The study also investigates the role of Principal Component Analysis (PCA) improves model efficiency by reducing the dimensionality of the feature set. Experimental results demonstrate that the integration of ML algorithms significantly improves IDS accuracy while reducing false alarms. This research offers valuable insights into addressing key IDS limitations and provides a comprehensive performance comparison to identify the most suitable model for real-world application.