Automted Fake News Prediction

docs/ML/projects/Automated_fakenews_detection.md

@@ -0,0 +1,179 @@
+# Fake News Detection Project  
+---
+
+### AIM
+
+To build a machine learning model that classifies news articles as either *True* or *Fake*, based on their textual content.
+
+---
+
+### DATASET LINK  
+
+[Dataset](https://www.kaggle.com/datasets/jainpooja/fake-news-detection)
+
+---
+
+### NOTEBOOK LINK  
+
+[Fake news Predictor Notebook](https://www.kaggle.com/code/disha520/fake-new-prediction-model)  
+
+---
+
+### LIBRARIES NEEDED  
+
+- pandas  
+- numpy  
+- scikit-learn  
+- matplotlib  
+- seaborn  
+- nltk  
+
+---
+
+### DESCRIPTION  
+
+Fake news has become a pressing issue in today's information-driven world. Detecting and categorizing news as genuine or fake can help improve the quality of content dissemination, particularly in critical areas like politics, health, and security.
+
+#### **Project Requirements**
+- Address the increasing spread of misinformation.
+- Provide reliable, automated methods for classification.
+- Train models on structured datasets to improve prediction accuracy.
+
+#### **Necessity**
+- Vital to curb misinformation spread.
+- Relevant to fields such as media, education, and cybersecurity.
+
+#### **Approach**
+- Using machine learning and natural language processing (NLP) techniques to extract meaningful features from text data.
+- Evaluate various models to find the best one for the classification task.
+
+#### **Additional Resources**
+- Blogs: "Intro to Fake News Detection Using NLP."
+- Research: “Detecting Fake News Using Machine Learning” (Journal XYZ, 2023).  
+
+---
+
+### EXPLANATION  
+
+#### Features Used:
+- **Text**: The main body of the article for feature extraction.
+- **Class**: Target variable indicating if an article is `Fake` or `True`.  
+
+---
+# Project Workflow
+
+## Step 1: **Importing Libraries**
+- Load essential libraries: `pandas`, `numpy`, `nltk`, `matplotlib`, `seaborn`, and machine learning libraries from `sklearn`.
+
+---
+
+## Step 2: **Downloading and Setting Up NLTK Resources**
+- Download stopwords and tokenizers for text preprocessing.
+
+---
+
+## Step 3: **Dataset Importing and Initial Processing**
+- Import datasets (`True.csv` and `Fake.csv`).
+- Add a `class` column to distinguish true (`1`) and fake (`0`) news.
+- Merge datasets into one combined `DataFrame`.
+- Drop unnecessary columns like `date` and `title`.
+
+---
+
+## Step 4: **Preprocessing Data**
+- Shuffle the combined dataset for randomness.
+- Define a function to clean and preprocess text:
+  - Tokenize and lowercase.
+  - Remove stopwords and non-alphanumeric tokens.
+- Apply the cleaning function to the `text` column.
+
+---
+
+## Step 5: **Splitting Data**
+- Divide data into features (`X`: processed text) and target (`y`: class).
+- Split into training and testing datasets (80% train, 20% test).
+
+---
+
+## Step 6: **Text Vectorization**
+- Convert text data to numerical form using `TF-IDF Vectorizer`:
+  - Extract a maximum of 5000 features.
+  - Fit on training data, transform both training and testing sets.
+
+---
+
+## Step 7: **Model Training and Evaluation**
+1. **Logistic Regression**:
+   - Train on `X_train_tfidf` and evaluate using accuracy, confusion matrix, and classification report.
+
+2. **Decision Tree Classifier**:
+   - Train the classifier and evaluate with the same metrics.
+
+3. **Gradient Boosting Classifier**:
+   - Apply boosting and assess model performance.
+
+4. **Random Forest Classifier**:
+   - Fit the ensemble model and evaluate.
+
+---
+
+## Step 8: **Visualization**
+1. **Accuracy Comparison**:
+   - Compare accuracy across models with a bar chart.
+
+2. **Correlation Matrix**:
+   - Display the correlation of predictions across all models in a heatmap.
+
+3. **Confusion Matrix**:
+   - Visualize confusion matrices for individual models.
+
+---
+
+### PROJECT TRADE-OFFS AND SOLUTIONS
+
+#### Trade-off 2: Interpretability vs. Complexity
+- **Issue:** Decision Trees are relatively easy to interpret, and their decision-making process is more transparent, but they tend to overfit and perform less effectively on more complex datasets (such as text data).
+- **Solution:** The challenge was addressed by using more powerful ensemble models like **Random Forests** and **Gradient Boosting**. While these models are less interpretable, they significantly improve accuracy due to their robust handling of more complex datasets. By combining multiple decision trees, **Random Forest** reduces overfitting and enhances predictive performance, offering a trade-off between reduced interpretability and better accuracy.
+
+#### Trade-off 3: Model Performance vs. Resource Consumption
+- **Issue:** Some models like **Gradient Boosting** and **Random Forest** are more resource-intensive and demand more computational power due to the complexity of their algorithms and the need to create multiple trees.
+- **Solution:** A variety of simpler models, such as **Logistic Regression** and **Decision Trees**, were initially evaluated to set performance benchmarks before running more complex and resource-heavy models. This helped balance between speed, resource usage, and performance, with regular evaluations of model accuracy. By comparing performance metrics, the trade-off between accuracy and resource consumption was managed efficiently.
+
+---
+
+### SCREENSHOTS  
+
+#### Workflow Overview  
+
+```mermaid
+graph LR
+  A[Import Libraries] --> B[Download NLTK Resources];
+  B --> C[Import Dataset];
+  C --> D[Preprocess Text Data];
+  D --> E[Vectorize Text];
+  E --> F[Train Models];
+  F --> G[Evaluate Models];
+  G --> H[Visualize Results];
+```
+
+### MODELS USED AND THEIR EVALUATION METRICS  
+
+| Model               | Accuracy | Precision | Recall |
+|---------------------|----------|-----------|--------|
+| Logistic Regression | 98.7%    | 98.3%     | 98.9%  |
+| Decision Tree       | 99.5%    | 99.7%     | 99.2%  |
+| Gradient Boosting   | 99.4%    | 99.0%     | 99.7%  |
+| Random Forest       | 99.7%    | 99.6%     | 99.7%  |
+
+---
+### CONCLUSION  
+
+#### Key Learnings  
+- **TF-IDF Vectorizer**: Effective for transforming text data into numerical features, crucial for building reliable models for text classification.  
+- **Text Preprocessing**: Cleaned and tokenized text data, removing stopwords and non-alphanumeric characters, which significantly improved model performance.  
+- **Ensemble Models**: Using more powerful models like **Random Forest** and **Gradient Boosting** provides higher accuracy compared to simpler models like **Logistic Regression** or **Decision Trees**, particularly when dealing with complex datasets like text.  
+
+
+#### Use Cases
+1. **News Verification Platforms**: Integrate the model into platforms like Fact-Check.  
+2. **Social Media Filtering**: Automatic classification of news shared across platforms.