BEHAVIOURAL RISK CLASSIFIER: MACHINE LEARNING ALGORITHMS TO CLASSIFY USERS BASED ON ONLINE BEHAVIOUR FOR IDENTIFYING POTENTIAL RISKS

Abstract

The rapid growth of online activity has significantly increased cybersecurity risks, as users often engage in behaviors that expose them to data breaches, identity theft, and other digital threats. Conventional security mechanisms—such as firewalls, intrusion detection systems, and antivirus software—primarily focus on known threats and operate reactively, limiting their ability to adapt to evolving user behavior. Moreover, traditional detection approaches rely on static thresholds and signature-based techniques, which are insufficient for identifying subtle or emerging behavioral risks. To address these challenges, this research proposes a Behavioral Risk Classifier, a machine learning– based framework designed to analyze user online behavior and classify users into safe or risky categories. The system utilizes a comprehensive set of behavioral features, including device type, social media usage, geolocation, network type, and e-safety awareness indicators. Data preprocessing involves timestamp decomposition, categorical encoding, feature scaling, and dimensionality reduction using Principal Component Analysis (PCA). To mitigate class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) is applied. The framework evaluates two models: a Gradient Boosting Classifier (GBC) and a hybrid Deep Neural Network–Random Forest Classifier (DNN–RFC), combining deep feature learning with robust ensemble prediction. Experimental results demonstrate accurate and proactive risk classification, enabling real-time behavioral risk assessment and enhanced cybersecurity decision-making.