In the modern era, cybercrime represents one of the most significant threats to global security, causing substantial financial, reputational, and operational damage to individuals, organizations, and governments. As cyber-attacks grow in complexity and frequency, conventional security mechanisms are increasingly inadequate to handle emerging threats. Consequently, the integration of machine learning in cybersecurity has gained considerable attention for its potential to provide intelligent, adaptive, and predictive solutions to combat evolving attack vectors. This paper investigates the application of supervised and unsupervised machine learning techniques to detect, forecast, and mitigate cyber-attacks. A comprehensive review of existing methodologies highlights the strengths and limitations of traditional intrusion detection systems and contrasts them with modern ML-based approaches. The proposed system employs a hybrid ensemble model combining Support Vector Machines (SVM) and Logistic Regression to classify various types of attacks and profile potential offenders based on demographic and behavioral attributes. An extensive dataset encompassing real-world network traffic and synthetic attack scenarios is used to train and validate the models. Data preprocessing includes feature extraction, normalization, and balancing to address class imbalance issues, which are common in cybersecurity datasets.
