An Improved Machine Learning Model to Enhance the Network Detection Rate

Abstract

Nowadays, the cybersecurity posture is getting worse than before because of the evolving attacks, and this paper presents an advanced adaptive weighted ensemble learning model based on machine learning aims to improving network security detection rates. The proposed model combines three base classifiers, Decision Tree, Support Vector Machine (SVM), and K-Nearest Neighbors (KNN). Optimize detection performance across various attack types by dynamically adjusts their weights based on real-time error and false alarm rates. In the experiments, the model achieved an accuracy of 97.8%, recall of 95.6%, and an F1 score of 96.2%, compared with traditional ensemble methods such as Random Forest and AdaBoost, which showed accuracies of 93.2% and 91.5%, respectively. Especially, the model maintained a false alarm rate of only 1.4%, significantly lower than that of the benchmark models, demonstrating its superior precision in classifying benign and malicious traffic. To further enhance model efficiency and accuracy, we applied the Boruta algorithm and Recursive Feature Elimination (RFE) for feature selection, which contributed to the elimination of noisy features and improved computational efficiency. Additionally, Principal Component Analysis (PCA) reduced dimensionality, decreasing model complexity while preserving essential data characteristics. These metrics like accuracy, recall and F1 score provide a comprehensive evaluation of the model's performance in terms of detecting malicious activities. Compared to traditional methods, the proposed model's adaptive weighting mechanism and feature selection processes ensure its robustness and flexibility across various network threat scenarios. These improvements provide more ideas for industries seeking to enhance their network security.