Abstract:

In today’s digital age, spam messages are a major nuisance, wasting user time and posing potential security threats. This project aims to develop a robust SMS spam detection system using Machine Learning techniques. We utilize three prominent classification algorithms — Naïve Bayes, Logistic Regression, and Support Vector Machine (SVM) — to classify text messages as either ‘spam’ or ‘ham’ (not spam). The system is trained on labeled SMS datasets and evaluated based on accuracy, precision, recall, and F1-score. The final goal is to integrate this system into applications like messaging platforms to filter out unwanted messages automatically.

Introduction:

SMS (Short Message Service) is a widely used communication medium, but it is increasingly plagued by spam messages. These messages can be promotional, fraudulent, or malicious. Manual filtering is inefficient and unreliable. Machine Learning (ML) offers a scalable and automated approach to detect such messages. This project explores three effective classification techniques—Naïve Bayes, Logistic Regression, and SVM—to build an SMS spam detection model that is both accurate and efficient.

Problem Statement:

With the rise in the number of spam SMS messages, users are facing serious issues related to privacy, financial fraud, and inconvenience. Traditional methods like keyword filtering are no longer sufficient due to the dynamic nature of spam. There is a need for an intelligent system that can automatically detect and filter spam messages with high accuracy.

Existing System and Disadvantages:

Existing System:

• Manual keyword-based filters.
• Rule-based spam detection systems.
• Basic spam filters used by mobile operating systems.

Disadvantages:

• Poor accuracy and adaptability.
• High false positives and negatives.
• Inability to detect intelligently crafted spam.
• No learning mechanism to adapt to new patterns.

Proposed System and Advantages:

Proposed System:

This project proposes an ML-based system that uses supervised learning techniques to classify SMS messages as spam or ham. It preprocesses text, extracts features (using TF-IDF or CountVectorizer), and then applies classification algorithms—Naïve Bayes, Logistic Regression, and SVM—to detect spam.

Advantages:

• High accuracy and precision.
• Can adapt and learn from new data.
• Capable of handling large volumes of SMS messages.
• Reduced false positive and false negative rates.
• Real-time and automated detection.

Modules:

1. Data Collection and Pre-processing
   • Clean SMS text, remove stopwords, stemming, tokenization.
2. Feature Extraction
   • Convert text into numerical features using CountVectorizer or TF-IDF.
3. Model Training
   • Train models using Naïve Bayes, Logistic Regression, and SVM.
4. Model Evaluation
   • Evaluate performance using metrics like accuracy, precision, recall, F1-score.
5. Prediction Module
   • Predict whether new SMS is spam or ham.
6. Web Interface 
   • Interface for users to test SMS messages.

Algorithms Used:

1. Naïve Bayes Classifier
   • Based on Bayes’ theorem with strong (naïve) independence assumptions.
2. Logistic Regression
   • Predicts the probability of class membership using sigmoid function.
3. Support Vector Machine (SVM)
   • Finds the optimal hyperplane that separates spam and ham classes.

Software and Hardware Requirements:

Software Requirements:

• Python 3.x
• Jupyter Notebook / VS Code / PyCharm
• Scikit-learn
• Pandas, Numpy, Matplotlib
• NLTK / spaCy for NLP
• Flask (for web interface, optional)

Hardware Requirements:

• Minimum 4GB RAM
• Intel i3 Processor or higher
• Internet connectivity for data download and testing

Conclusion:

This project successfully demonstrates how machine learning techniques like Naïve Bayes, Logistic Regression, and SVM can be applied to detect spam messages effectively. Among the models, each has its strengths in terms of speed, accuracy, and interpretability. The proposed system provides a scalable and automated solution to a widespread problem in mobile communication.

Future Enhancement:

• Deploy the system in real-time messaging apps.
• Use Deep Learning models like LSTM or BERT for better accuracy.
• Multilingual SMS spam detection.
• Continual learning using feedback from users.
• Integration with email or social media platforms for spam detection.