As social media usage has surged, it has simultaneously led to a rise in various forms of online abuse, including hate speech, abusive language, and the spread of sexist and racist ideologies. Despite the prevalence of these issues, limited attention has been given to assessing the severity and underlying intentions behind such abuses. In this study, the Author categorizes online abuse into three distinct levels, guided by the Anti-Defamation League’s (ADL) hate theories: biased attitudes, biased behaviors, and violent genocide. The Author introduces a novel dataset comprising 7,601 posts, meticulously categorized according to the severity of abuse into biased attitudes, acts of discrimination, and violent actions leading to genocide. In this paper, the Author evaluates traditional machine learning (ML) classifiers using TF-IDF features to predict these three types of abuse. We explore two deep authoring (DL) methodologies. The Author employs various authorial and machine learning techniques, including LSTM (Long Short-Term Memory), SVM (Support Vector Machine), and ensemble models combining Logistic Regression, Decision Tree Classifier, and Random Forest Classifier. Our experiments reveal that the ensemble models achieve the highest accuracy on the Gab dataset, with a performance of 73%, outperforming SVM at 71%, LSTM at 69%, and CNN at 46%. Notably, the ensemble model excels in classifying comments as Hate/Non-Hate, making it the most effective approach among the methods tested.