News Topic Classification Using Naive Bayes and Logistic Regression

Overview

This project classifies news articles into four categories: World News, Sports, Business, and Tech.
The dataset includes:

• Training Data: 120,000 samples
• Testing Data: 7,600 samples
  Both datasets are balanced across the categories.

The goal is to preprocess the text data, extract features, and train models using Naive Bayes and Logistic Regression to achieve high classification performance.

---

Features

Text Preprocessing

• Removal of HTML tags and URLs.
• Tokenization, lemmatization, and stopwords removal.
• Custom filtering of extra common words.

Visualization

• Word clouds for overall and category-specific analysis.

Feature Engineering

• TF-IDF vectorization with tri-grams and feature selection.

Model Training

• Hyperparameter tuning using GridSearchCV.
• Evaluation of models using precision, recall, and F1-score.

Error Analysis

• Insights into misclassifications and overlapping categories.

---

Logistic Regression

• Implemented in a scikit-learn Pipeline for ease of use.

---

Results

Metric	Testing Data
Precision	85.00%
Recall	84.82%
F1-Score	84.90%

Observations

• Misclassifications often occur between Business and Tech due to overlapping content.

Installation

Clone the Repository

git clone https://github.com/yourusername/news-topic-classification.git
cd news-topic-classification