anu56787ty/spam-detector-tfidf-lr

📧 Spam Detector — TF-IDF + Logistic Regression

A lightweight, free spam email/SMS classifier trained on the SMS Spam Collection dataset.

Zero cost to run — no GPU needed, CPU-only inference.

📊 Performance

Metric	Score
Accuracy	98.65%
F1 Score (spam)	94.88%
Precision (spam)	96.53%
Recall (spam)	93.29%

Evaluated on 1,115 held-out test messages (20% split).

Confusion Matrix:

              Predicted Ham   Predicted Spam
Actual Ham         961              5
Actual Spam         10            139

Only 10 false negatives (spam missed) and 5 false positives out of 1,115 test samples.

🗂️ Dataset

• Source: ucirvine/sms_spam
• Size: 5,574 SMS messages (4,827 ham + 747 spam)
• Split: 80% train / 20% test, stratified

🚀 Quick Start

import pickle
from huggingface_hub import hf_hub_download

# Load model
tfidf_path = hf_hub_download("anu56787ty/spam-detector-tfidf-lr", "tfidf_vectorizer.pkl")
clf_path   = hf_hub_download("anu56787ty/spam-detector-tfidf-lr", "logistic_regression.pkl")

with open(tfidf_path, "rb") as f:
    tfidf = pickle.load(f)
with open(clf_path, "rb") as f:
    clf = pickle.load(f)

def predict(text):
    vec   = tfidf.transform([text])
    pred  = clf.predict(vec)[0]
    proba = clf.predict_proba(vec)[0]
    return {"label": "spam" if pred == 1 else "ham", "confidence": proba[pred]}

# Try it!
print(predict("Congratulations! You won a FREE iPhone! Click now!"))
# → {'label': 'spam', 'confidence': 0.977}

print(predict("Hey, what time are we meeting for lunch?"))
# → {'label': 'ham', 'confidence': 0.986}

🏗️ Architecture

Input text
    ↓
TF-IDF Vectorizer
  • max_features=10,000
  • ngram_range=(1,2)   ← unigrams + bigrams
  • sublinear_tf=True
    ↓
Logistic Regression
  • C=5.0
  • class_weight='balanced'   ← handles class imbalance
    ↓
Output: ham / spam + confidence score

✅ Why This Model?

Feature	Value
💰 Cost	Free — $0
⚡ Speed	< 1ms per prediction
💾 Size	~2 MB total
🖥️ Hardware	CPU only
📦 Dependencies	scikit-learn, huggingface_hub

📈 Training

Dataset : ucirvine/sms_spam (5,574 messages)
Train   : 4,459 messages
Test    : 1,115 messages
Time    : < 5 seconds on CPU

Generated by ML Intern

This model repository was generated by ML Intern, an agent for machine learning research and development on the Hugging Face Hub.

• Try ML Intern: https://smolagents-ml-intern.hf.space
• Source code: https://github.com/huggingface/ml-intern

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = 'anu56787ty/spam-detector-tfidf-lr'
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

For non-causal architectures, replace AutoModelForCausalLM with the appropriate AutoModel class.