Detection of Brute-force Login Attempts Using Machine Learning

Abstract

Identifying brute-force and dictionary-based login attempts in modern cybersecurity systems has become increasingly challenging, as advanced techniques often fail to detect large-scale intrusion attempts.The aim of this research is to determine the effectiveness of machine learning methods in identifying such attacks in an accurately and efficiently. Two classifiers SVM and GNB, are trained on authentication log data, both with and without PCA for dimensionality reduction.The experimental results indicate that SVM achieves the highest accuracy of 97.24% without PCA and 96.55% with PCA, demonstrating that SVM is robust in high-dimensional feature spaces. Conversely, GNB shows significant with PCA, with accuracy rising from 87.93% to 91.03%, highlighting the importance of feature decorrelation in probabilistic models. The key contribution of this work is the comparative study of lightweight machine learning models demonstrating that PCA improves the performance of correlation- sensitive classifiers without undermining the computational efficiency. The results provide a feasible and scalable solution to real-time intrusion detection systems.