Twitter Sentiment-Analysis

Overview

This repository contains a Twitter sentiment analysis project using classical machine learning models. The goal is to classify tweets into four sentiment categories: Positive, Negative, Neutral, and Irrelevant. The project includes full preprocessing, feature extraction, model training, evaluation, and saving the final model for deployment.

---

Dataset

The dataset used for this project is taken from Kaggle: Twitter Entity Sentiment Analysis.

• Training set: twitter_training.csv
• Validation set: twitter_validation.csv
• Columns:
  • ID → Tweet ID
  • Topic → Topic of the tweet
  • Sentiment → Sentiment label (Positive, Negative, Neutral, Irrelevant)
  • Text → Original tweet text

---

Features & Preprocessing

1. Text Cleaning

   • Lowercasing, HTML decoding
   • Remove URLs, mentions (@user), hashtags, emojis, and special characters
   • Tokenization, stopwords removal, lemmatization

2. Label Encoding

   • Convert sentiment labels into integers for model training

3. Feature Extraction

   • TF-IDF vectorization with unigrams and bigrams
   • Max features: 10,000

---

Models Used

• Logistic Regression
• Multinomial Naive Bayes
• Decision Tree
• Random Forest (final model)
• Linear SVM

Evaluation metrics: Accuracy, Precision, Recall, F1-score, Confusion Matrix, ROC Curve

Final Model: Random Forest with 200 estimators trained on the full dataset.

---

Results

The table below shows the accuracy of different machine learning models on the validation set:

Model	Accuracy
Logistic Regression	0.865
Naive Bayes	0.745
Decision Tree	0.886
Random Forest	0.946
Linear SVM	0.904

Key Insight:

• Random Forest achieved the highest accuracy (0.946) across all classes.
• It demonstrates the best balance between precision, recall, and F1-score.
• Therefore, Random Forest is selected as the final model for deployment and further predictions.

---