Machine Learning methods for Online Toxic Comment Classification

Abstract

With the rapid expansion of social media and online communication platforms, the proliferation of toxic comments such as hate speech, harassment, and offensive language—has become a significant concern. Detecting and moderating such content is crucial for maintaining healthy digital interactions. This paper presents an overview of machine learning (ML) methods used for online toxic comment classification. Various supervised and deep learning approaches, including Logistic Regression, Support Vector Machines (SVM), Random Forests, Recurrent Neural Networks (RNN), and Convolutional Neural Networks (CNN), are analyzed for their performance in identifying toxic behavior in textual data. The study emphasizes the importance of data preprocessing, feature extraction techniques (such as TF-IDF and word embeddings), and model optimization for improving classification accuracy. It also highlights recent advancements in transformer-based models like BERT, which achieve superior context understanding and robustness. Challenges such as data imbalance, context ambiguity, and evolving language patterns are discussed. The paper concludes that integrating traditional ML with deep learning and Natural Language Processing (NLP) techniques offers a powerful solution for effective and scalable toxic comment detection.