metadata
language: en
license: apache-2.0
tags:
- text-classification
- spam-detection
- distilbert
- sms
datasets:
- sms_spam
metrics:
- accuracy
- f1
- precision
- recall
widget:
- text: Congratulations! You've won a FREE prize. Call now!
- text: Hey, are we still meeting for lunch tomorrow?
- text: 'URGENT: Your account has been compromised!'
- text: Can you pick up milk on the way home?
SMS Spam Detector
Model Description
Fine-tuned DistilBERT model for detecting spam SMS messages.
Performance:
- Accuracy: 0.9916 (99.16%)
- Precision: 0.9730
- Recall: 0.9643
- F1-Score: 0.9686
- ROC-AUC: 0.9990
Quick Start
from transformers import pipeline
classifier = pipeline("text-classification", model="didulantha/sms-spam-detector")
result = classifier("Win a free prize now!")
print(result)
Detailed Usage
from transformers import DistilBertTokenizer, DistilBertForSequenceClassification
import torch
tokenizer = DistilBertTokenizer.from_pretrained("didulantha/sms-spam-detector")
model = DistilBertForSequenceClassification.from_pretrained("didulantha/sms-spam-detector")
def predict(text):
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=128)
outputs = model(**inputs)
probs = torch.softmax(outputs.logits, dim=1)
return "SPAM" if probs[0][1] > 0.5 else "HAM"
print(predict("Free prize!"))
Training Data
- Dataset: SMS Spam Collection (5,574 messages)
- Train/Val/Test split: 70/15/15
- Ham: 86.6%, Spam: 13.4%
Training Details
- Base model: distilbert-base-uncased
- Epochs: 3
- Batch size: 16
- Learning rate: 2e-05
- Optimizer: AdamW
Limitations
- Trained on English SMS messages only
- May not generalize well to other languages
- Performance may degrade on heavily obfuscated spam
Citation
@misc{sms-spam-detector,
author = {Isuru Didulantha},
title = {SMS Spam Detector},
year = {2025},
publisher = {HuggingFace},
url = {https://huggingface.co/didulantha/sms-spam-detector}
}