Advanced-Toxic-Comment-Classifier

📜 Project Description

The Toxic Comment Classifier is an Advanced machine-learning project developed to detect and categorize toxic content in user comments. By employing a multi-label classification strategy, it addresses the critical challenges faced by online platforms in identifying harmful language. Designed and implemented in a Jupyter Notebook environment, the project utilizes cutting-edge techniques for feature extraction and classification. This advanced solution is deployed for real-time toxic comment detection, leveraging Convolutional Neural Networks (CNNs) and a Kaggle dataset to achieve an impressive accuracy of 95.31%. The workflow encompasses comprehensive preprocessing, feature extraction, and detailed evaluation, enabling effective identification of hate speech, harassment, and offensive language.

🎯Objectives

Build a robust classifier to accurately identify toxic comments across six predefined categories:

• Toxic
• Severe Toxic
• Obscene
• Threat
• Insult
• Identity Hate Applying advanced preprocessing techniques and machine learning techniques to ensure high accuracy. Additionally, analyze and provide insights into the patterns and distribution of toxic content within the dataset.

🛠️ Tools and Technologies

• Programming Language: Python
• Development Environment: Jupyter Notebook
• Libraries: Pandas, NumPy, Matplotlib, Seaborn, Scikit-learn, TensorFlow
• Visualization Tools: Matplotlib and Seaborn for in-depth exploratory data analysis (EDA).

🚀 Methodology

The model was built using Convolutional Neural Networks (CNN) to detect toxic comments across six categories.

• Data Preprocessing: Text was cleaned, tokenized, and padded to ensure uniform input for the CNN model.
• Feature Extraction: An embedding layer was used to represent words as dense vectors, capturing semantic relationships.
• Model Architecture: The CNN model consists of convolutional layers to extract features, followed by max-pooling for dimensionality reduction and a fully connected layer for classification.
• Training: The model was trained using the Adam optimizer and categorical cross-entropy loss function.
• Evaluation: The model achieved 95.31% accuracy with additional metrics like precision, recall, and F1-score for balanced performance across all categories.

📈Results

The model’s performance was evaluated using multiple metrics to ensure high accuracy and reliability in detecting toxic comments across six categories. The key evaluation metrics are as follows:

• Accuracy: 95.31%
• Precision: 95.08%
• Recall: 95.31%
• F1-Score: 94.92%

Confusion Matrix A Confusion Matrix was created to visually represent the model’s performance in classifying the toxic comments. The matrix shows how well the model distinguishes between the six predefined categories: Toxic, Severe Toxic, Obscene, Threat, Insult, and Identity Hate. Confusion Matrix :(https://github.com/yashisingh-ds/Advanced-Toxic-Comment-Classifier/blob/main/confusionmatrix.png)

Model Performance Visualizations

The following performance visualizations provide deeper insights into the model’s training and evaluation:

• Learning Curve for Training Accuracy: Displays how the model's accuracy improves over time during training. (https://github.com/yashisingh-ds/Advanced-Toxic-Comment-Classifier/blob/main/LearningCurve(Training%20Accuracy).png)
• Learning Curve for Validation Accuracy: Shows how the model performs on validation data across training epochs. (https://github.com/yashisingh-ds/Advanced-Toxic-Comment-Classifier/blob/main/LearningCurve(Validation%20Accuracy).png)
• ROC Curve: Illustrates the model's ability to distinguish between classes, with a higher curve indicating better performance.(https://github.com/yashisingh-ds/Advanced-Toxic-Comment-Classifier/blob/main/ROC%20Curve.png)
• Precision-Recall Curve: Highlights the trade-off between precision and recall, helping evaluate the model's effectiveness in handling imbalanced data. (https://github.com/yashisingh-ds/Advanced-Toxic-Comment-Classifier/blob/main/Precison-Recall%20Curve.png)
• ROC Curve Comparison: Compares the performance of the model using different metrics and thresholds, offering a holistic view of its classification capabilities.( https://github.com/yashisingh-ds/Advanced-Toxic-Comment-Classifier/blob/main/ROC%20Curve%20Comparison.png)

📊 Conclusion

The Toxic Comment Classifier achieved a high accuracy of 95.31% in detecting harmful content across multiple categories. The model performs well in real-world scenarios, and future improvements can further enhance its effectiveness and scalability.