SMS Spam Detection Model Using NLP

Goal of the Model

The main objective of this project is to classify SMS messages as Spam or Ham (not spam) using various Natural Language Processing (NLP) techniques and machine learning algorithms.

Source of Data

The dataset used for this project is sourced from Kaggle.

---

Work Flow

1. Importing the required libraries
2. Loading the Dataset
3. Basic Understanding of Data
4. Data Preprocessing
5. Exploratory Data Analysis (EDA) & Text Preprocessing
6. Model Training, Building, and Evaluation

---

Key Tasks in Text Preprocessing Using Natural Language Tool Kit Library (NLTK) :

• Lowercasing all text data
• Tokenization: Splitting text into individual words
• Removing special characters, stopwords, and punctuation
• Stemming: Reducing words to their root form
• Finding most common words in Ham and Spam messages
• Analyzing 30 most frequent words in Ham and Spam messages

Feature Engineering in EDA:

• Calculated the number of characters, words, and sentences in each message
• Performed Univariate/Bivariate Analysis to explore the dataset
• SMOTE technique was applied to handle the imbalance in the dataset

---

Modeling and Algorithms

We trained, built, and evaluated the model using the following algorithms:

• Naive Bayes
• Support Vector Machine
• K-Nearest Neighbors
• Decision Tree
• Logistic Regression
• Random Forest
• AdaBoost
• Bagging Classifier
• Extra Trees Classifier
• Gradient Boosting
• XGBoost

Additionally, applied for improving the performance

• Voting Classifier
• Stacking Classifier

Model Performance Evaluation

Model Performance Observation

• Cross-Validation Accuracy:

  • The average accuracy from cross-validation is 99.85%, with minimal variation, which indicates high consistency in performance.

• Training Accuracy:

  • The model achieved a perfect 100% accuracy on the training data, showcasing its ability to learn the data effectively.

• Test Accuracy:

  • The test accuracy is 99.89%, confirming that the model generalizes well on unseen data.

• Precision & Recall:

  • Both precision and recall are 99.89%, meaning the model is highly accurate in predicting spam and non-spam messages.

Conclusion:

The model shows excellent performance with high accuracy, precision, and recall, making it a robust solution for SMS Spam Detection.