Millions of comments are written by users every day on different posts as social media has gained immense popularity. But abusive language used in these types of interactions is at an all-time high. Previous studies have predominantly explored either traditional ML or DL models in isolation, often evaluating performance using basic metrics such as accuracy and F1-score. This study is concerned with filtering abusive language to encourage respectful conversation by examining several machine learning models such as XGBoost, Support Vector Machine (SVM), and Logistic Regression, along with hyperparameter tuning and deep learning models such as Dynamic Memory Network (DMN), Hierarchical Attention Network (HAN), and Generative Adversarial Network (GAN). The performance of the models is evaluated and compared with advanced metrics such as MCC, specificity, Cohen’s Kappa, Hamming loss, and GMean along with precision, recall, and f1-score. The results provide valuable insights regarding the strengths and weaknesses of each model and demonstrate that an ensemble of model types can create a powerful classification for abusive language.