The dissemination of harmful information and hate speech has become a significant problem on social media and other internet platforms. The quick spread of such toxicity can cause severe damage to people’s mental health. Therefore, the need for an intelligent system capable of detecting and classifying hate speech in real time has become more pressing than ever. This paper aims to devise an effective feature selection algorithm for detecting and classifying toxicity on the internet which can prevent the proliferation of hateful content on social media. To achieve this objective, we propose a novel feature engineering technique called HateSwarm. This technique employs bio-inspired algorithms to select efficient features for the binary classification of non-hate and hate speech. We train the baseline machine learning models on these features and evaluate the proposed algorithm’s performance on two benchmark datasets. The proposed algorithm achieves a remarkable classification accuracy of 92% and competes favorably with state-of-the-art models. Experimental results demonstrate that our approach is effective in detecting and preventing hate speech on the internet. This paper presents a novel approach to combat the growing problem of harmful information and hate speech on social media and other internet platforms.