ABSTRACT
The rise of social media platforms and user-generated content has led to an increase in online hate speech, offensive language, and toxic behavior. To address this growing issue, we propose a machine learning-based web application capable of classifying user-submitted text into predefined categories: hate speech, offensive language, or neither. This system leverages natural language processing (NLP) techniques and supervised machine learning models to provide a robust and scalable solution for hate speech detection. The core engine of the system uses a Logistic Regression classifier, trained on a real-world labeled dataset containing tweets annotated for hatefulness. Prior to model training, extensive data preprocessing is performed, including lowercasing, URL and punctuation removal, number filtering, and stopword elimination using the NLTK library. Text data is then transformed into numerical vectors using TF-IDF vectorization with unigrams and bigrams to capture contextual information. To tackle the issue of class imbalance in the dataset, Synthetic Minority Over-sampling Technique (SMOTE) is applied. The balanced dataset is then used to train a Logistic Regression model, which demonstrates high accuracy and reliability. The trained model, along with the fitted TF-IDF vectorizer, is serialized using joblib for deployment.
The frontend of the application is built using Flask and HTML templates, providing users with a seamless interface to register, log in, and input text for real-time analysis. The system securely stores user data in a MySQL database, ensuring authentication and session management. Passwords are hashed using the Werkzeug library for added security. Upon submission, the input text undergoes cleaning, vectorization, and classification, and the result is displayed to the user. This web-based platform effectively combines the strengths of data science and web development to create a practical tool for moderating online content. It can be extended for social media monitoring, community moderation, or integration into broader content filtering systems.
Overall, the project demonstrates the practical application of machine learning techniques in a real-world setting, showcasing how intelligent systems can help mitigate digital abuse and promote safer online communities.


OBJECTIVES
 To develop a machine learning-based classification system capable of accurately detecting and distinguishing between hate speech, offensive language, and neutral content in social media text using techniques like TF-IDF and Logistic Regression.
 To preprocess raw textual data effectively by implementing standard NLP techniques such as stopword removal, lowercasing, punctuation cleaning, and text normalization for improved feature extraction and model performance.
 To handle class imbalance in the dataset using SMOTE (Synthetic Minority Over-sampling Technique), ensuring that the classifier performs fairly across all categories and does not favor majority classes.
 To build a secure, user-friendly Flask web interface that allows users to register, log in, input text, and view classification results in real time, promoting accessibility and practical deployment.
 To evaluate the performance of the model using appropriate metrics like accuracy, precision, recall, and F1-score, and visualize the confusion matrix for better interpretability and analysis.