Spam Detection using Word2Vec and Logistic Regression

This repository contains an implementation of a spam detection model using Word2Vec for text representation and Logistic Regression for classification. The model is trained to classify messages as spam or not spam based on their content.

Features

• Implements spam detection using a machine learning approach.
• Uses Word2Vec embeddings to convert text into numerical vectors.
• Applies Logistic Regression for binary classification.
• Utilizes gensim and scikit-learn for text processing and model training.
• Evaluates model performance using accuracy, precision, and recall.

Dataset

The project works with a dataset containing labeled messages:

• Spam messages: Unwanted or fraudulent messages.
• Non-spam messages: Normal, legitimate messages.
• Preprocessing steps include tokenization, stopword removal, and embedding conversion.

Model Architecture

Text Embedding using Word2Vec

• Tokenization & Preprocessing: Messages are cleaned and tokenized.
• Word2Vec Embeddings: Converts words into dense vector representations based on their semantic similarity.
• Feature Extraction: Sentence embeddings are generated by averaging word vectors.

Classification Model using Logistic Regression

• Input Features: Sentence-level Word2Vec embeddings.
• Logistic Regression Classifier: A simple yet effective model for binary classification.
• Probability Estimation: Uses the sigmoid function to predict the likelihood of a message being spam.

Training Process

1. Data Preprocessing:
   • Tokenization and text cleaning.
   • Training or loading pre-trained Word2Vec embeddings.
   • Averaging word vectors to represent sentences.
2. Model Training:
   • Logistic Regression trained with cross-entropy loss.
   • Optimization using Stochastic Gradient Descent (SGD) or other solvers.
3. Validation & Evaluation:
   • Accuracy, precision, recall, and F1-score for classification tasks.
   • Confusion matrix for performance analysis.

Evaluation

• Accuracy: Measures how well the classifier distinguishes spam from non-spam.
• Precision & Recall: Evaluates classification performance in detecting spam messages.
• Confusion Matrix: Visualizes model performance on test data.

Usage

• The model takes an input message and classifies it as spam or not spam.
• Can be integrated into email filtering, SMS classification, or chatbot moderation systems.