Hate Speech that attacks people or groups while discriminating along promoting violence exists because of race gender religious and ethnic attributes. Social media expansion has led to widespread growth of hate speech which now represents a severe risk for online security. The two traditional machine learning methods Support Vector Machines (SVM) and Naïve Bayes faced difficulties regarding dataset imbalance and context interpretation as well as feature recognition which produced subpar results during classification tasks and many incorrect negative classifications. This study implements Long Short-Term Memory (LSTM) and Convolutional Neural Networks with Bidirectional LSTM (CNN-BiLSTM) deep learning models to effectively process sequential text data dependencies and contextual relationships. Monitoring occurred on Twitter to acquire 31,962 tweets after which NLP methods were employed for tokenization and removal of stop-words followed by normalization and oversampling techniques to address dataset imbalance. Analysis results indicate that CNN-BiLSTM surpasses LSTM by reaching 96.2% accuracy and generating 86.2% precision along with 95.0% recall and finally delivering a 90.4% F1-score to establish its dominance in detecting hate speech. This study advances the detection of hate speech in real time which offers support to social media platforms when handling content moderation policies. The study should examine in future how transformer-based architectures along with multilingual datasets would enhance generalization abilities in future investigations.