Cyberbullying has become a tremendous issue in the information age, inflicting notable psychological and emotional harm on its victims. While machine learning (ML) and deep learning (DL) approaches have achieved notable progress in detecting cyberbullying, most studies have concentrated on Englishlanguage data, neglecting native languages like Hindi and other regional dialects. This review explores recent progress in automated hurtful sentiment detection, focusing on the integration of traditional (ML) methods and advanced deep learning (DL) frameworks and hybrid models. It places particular attention on challenges unique to the Hindi language, such as the scarcity of annotated datasets, linguistic complexities, and the frequent use of codeswitching. The effectiveness of methods like convolutional neural networks (CNN), long short-term memory (LSTM), and bidirectional LSTM (BiLSTM) is examined to understand their ability to identify patterns and circumstantial relationships in text. This research specifies cyberbullying comments using a multi-class classification strategy. Kaggle is used to collect the dataset, which is later used to train the model and evaluate it. The dataset contains 16,923 Hinglish/Hindi comments, consisting of four different classes: Non-Aggressive (NAG), Offensive-Aggressive (OAG), Contextually-Aggressive (CAG), and Gendered (GEN). In this study, the CNNBiLSTM model is implemented and achieves an accuracy of 96.10%. This paper aims to direct future research toward developing inclusive, precise, and scalable solutions for detecting cyberbullying in Hindi/Hinglish text and cultural landscapes.