Content Blocking Method To Reduce False Positives Based On Machine Learning

Abstract

This study presents an experimental approach to enhance content-blocking systems by integrating machine learning with domain classification and Pi-hole DNS server technology. While traditional blocking mechanisms often result in false positives—legitimate domains mistakenly blocked—this research aims to mitigate such issues. By implementing various testing scenarios, including TF-IDF and N-gram feature extraction with and without preprocessing, the study evaluates the classification performance using the Naive Bayes algorithm. The results reveal the highest accuracy of 84% achieved with the N-gram method without preprocessing. This integrated approach shows promise in improving the precision of ad and website blocking mechanisms.