📧 Spam Classifier with BOW, TF-IDF, and Word2Vec

This project aims to classify messages as Spam or Ham (Not Spam) using different text vectorization techniques:

• Bag of Words (BOW)
• TF-IDF (Term Frequency–Inverse Document Frequency)
• Word2Vec (Averaged Embeddings)

We compare the performance of these vectorization techniques using classical machine learning models and evaluate them using the AUC-ROC score.

---

🧠 Dataset

We use the SMS Spam Collection Dataset from the UCI repository.

• Total Samples: ~5,500
• Labels: ham, spam
• Text column: message

---

📊 Preprocessing Steps

1. Lowercasing
2. Removing punctuation and stopwords
3. Tokenization
4. Lemmatization
5. Train-test split (80-20)

---

🔧 Vectorization Techniques

✅ Bag of Words (CountVectorizer)

• Converts text to word frequency vectors.
• Sparse and high-dimensional.
• Works well with Naive Bayes and Logistic Regression.

✅ TF-IDF (TfidfVectorizer)

• Weights rare but informative words higher.
• Similar to BOW but more refined.
• Useful for downplaying common words.

✅ Word2Vec (Averaged)

• Self-trained word embeddings.
• Sentence represented as the average of word vectors.
• Dense, low-dimensional vectors,vector_size=100
• Loses context (e.g., "not good" = "good").

---

🚀 Models Used

• Logistic Regression
• All models trained separately on each vectorization output.

---

📈 Evaluation Metric

I use AUC-ROC (Area Under the Curve – Receiver Operating Characteristic) to evaluate model performance.

---

📊 Results

Vectorization	AUC-ROC Score
Bag of Words	0.98
TF-IDF	0.98
Word2Vec (avg)	0.78

Note: Word2Vec gives lower performance in this context due to loss of semantic structure when averaging and lack of context-awareness.

---