SMS Spam Classification

This project implements and compares different Natural Language Processing (NLP) techniques for SMS spam classification using various text vectorization methods and machine learning models.

Project Overview

The project explores three different approaches to SMS spam classification:

1. Bag of Words (BOW)
2. Term Frequency-Inverse Document Frequency (TF-IDF)
3. Word2Vec with Average Word Embeddings

Dataset

The project uses the SMS Spam Collection dataset, which contains labeled SMS messages categorized as either spam or ham (legitimate messages). The dataset is loaded from SMSSpamCollection with two columns:

• label: Indicates whether the message is spam or ham
• message: The actual text content of the SMS

Implementation Details

Common Preprocessing Steps

• Text cleaning using regular expressions
• Conversion to lowercase
• Removal of stopwords
• Text normalization (stemming/lemmatization)

1. Bag of Words (BOW) Approach

• Uses CountVectorizer from scikit-learn
• Features:
  • Maximum 2,500 features
  • Includes unigrams and bigrams (ngram_range=(1,2))
• Model: Multinomial Naive Bayes classifier

2. TF-IDF Approach

• Uses TfidfVectorizer from scikit-learn
• Features:
  • Maximum 2,500 features
  • Includes unigrams and bigrams (ngram_range=(1,2))
• Model: Multinomial Naive Bayes classifier

3. Word2Vec with Average Word Embeddings

• Uses gensim Word2Vec model
• Features:
  • Custom trained Word2Vec model on the SMS dataset
  • Average word vectors for sentence representation
  • Vector dimension standardization to 100 features
• Model: Random Forest Classifier

Project Structure

|
├── 1_Spam_Classification_using_BOW.ipynb        # BOW implementation
├── 2_Spam_Classification_using_TF-IDF.ipynb     # TF-IDF implementation
├── 3_Spam_Classification_using_Word2Vec_AvgWord2Vec.ipynb  # Word2Vec implementation
└── README.md

Requirements

The project requires the following Python libraries:

• pandas
• numpy
• scikit-learn
• nltk
• gensim
• tqdm

Model Performance

Each approach was evaluated using standard classification metrics including accuracy, precision, recall, and F1-score. The models were trained on 80% of the data and tested on the remaining 20%.

Usage

1. Install the required packages:

   pip install pandas numpy scikit-learn nltk gensim tqdm
   

2. Download required NLTK data:

   import nltk
   nltk.download('stopwords')
   nltk.download('wordnet')
   nltk.download('punkt_tab')
   

3. Run the Jupyter notebooks in sequence to compare different approaches:
   • Start with BOW implementation
   • Move to TF-IDF implementation
   • Finally, try the Word2Vec implementation