COOKIE CLASSIFICATION FRAMEWORK FOR ENHANCING USER PRIVACY CONSENT MANAGEMENT

ABSTRACT

Internet web cookies are vital to customization and analytics, notwithstanding most cookies allow behavioural profiling and cross-site tracking without the consent of web users, this has raised various privacy concerns within GDPR and related regulations. Various available consent management systems are built on static rule-base classification which has resulted in accuracies, opaqueness and unscalable. This study presents a framework for automated cookie classification that enhances privacy consent management. Playwright crawler was used to automate the collection of 59,926 cookie records from 10,000 websites in a pre- and post-consent visit process. Each cookie was engineered using 23 feature vector that capture temporal, structural, domain, security and behavioural properties. Four supervised classifiers were trained and evaluated; Random Forest, Logistic Regression, Support Vector Machine, Multilayer Perception. Random Forest achieved the highest performance in comparison with the other three with an accuracy of 98.83%, macro F1 of 0.9878, ROC-AUC of 0.9997 and Privacy-Risk Misclassification Rate of 1.01%. Unsupervised K-Means clustering of 21,866 unlabeled cookies revealed five behavioural grouping, which includes pre-consent violation cluster and persistent third-party tracking cluster, that demonstrate the frameworks capacity for proactive GDPR compliance monitoring beyond supervised classification.

Keywords: Web Tracking Detection, Machine Learning, Cookie Classification, Privacy Consent Management, GDPR Compliance, Explainable AI