linggarm / news-category-classification-using-tf-idf-vectorizer-and-multinomial-naive-bayes

News Category Classification using TF-IDF Vectorizer and Multinomial Naive Bayes with Hyperparameter Tuning

News-Category-Classification-using-TF-IDF-Vectorizer-and-Multinomial-Naive-Bayes

News Category Classification using TF-IDF Vectorizer and Multinomial Naive Bayes with Hyperparameter Tuning

About the Project

This project focuses on classifying news articles into different categories using a combination of TF-IDF Vectorizer as Word Embedding (to convert text data into vectors) and the Multinomial Naive Bayes algorithm to perform text classification. The dataset utilized is the 20newsgroups dataset, consisting of 18,846 samples of news data distributed across 20 classes. The headers and footers of each news article are included to reduce overfitting.

Technology Used

• Python
• Pandas
• Numpy
• Matplotlib
• Seaborn
• Scikit-learn

Notebook File

• notebooks/news_category_classification.ipynb

Workflow

Data Collection

• This project utilizes the 20newsgroups dataset, which consists of a total of 18,846 samples of news data, categorized into 20 classes.

• The 20newsgroups dataset is obtained from the Scikit Learn Dataset: sklearn.datasets.fetch_20newsgroups.

• The headers and footers of each news data are included in this project to reduce overfitting.

• The 20 classes present in the dataset are as follows:

  1. rec.sport.hockey
  2. soc.religion.christian
  3. rec.motorcycles
  4. rec.sport.baseball
  5. sci.crypt
  6. rec.autos
  7. sci.med
  8. comp.windows.x
  9. sci.space
  10. comp.os.ms-windows.misc
  11. sci.electronics
  12. comp.sys.ibm.pc.hardware
  13. misc.forsale
  14. comp.graphics
  15. comp.sys.mac.hardware
  16. talk.politics.mideast
  17. talk.politics.guns
  18. alt.atheism
  19. talk.politics.misc
  20. talk.religion.misc

• Here is the data distribution for each class:

  

Data Preprocessing

• The data preprocessing steps applied to the data include:
  • Case folding: Transform all letters into lowercase.
  • Stopword removal using TF-IDF Vectorizer provided by Scikit Learn library
  • Train the TF-IDF Vectorizer model using TF-IDF Vectorizer provided by Scikit Learn library
• The trained TF-IDF Vectorizer model is saved in models/news_vectorizer.sav

Data Splitting

• The data is split into training data and testing data with a ratio of 0.2, signifying 80% for training data and 20% for testing data. The random_state variable is set to 0.

Model Building & Training

• The model is trained using the Multinomial Naive Bayes algorithm.
• To obtain the best parameters, the training process is conducted using the Hyperparameter Tuning technique with the Grid Search method, which involves trying all possible combinations from a list of hyperparameters.
• Here is the list of hyperparameters used in the training process:
  • 'alpha' (Additive (Laplace/Lidstone) smoothing parameter) :
    • 0,
    • 1.0,
    • 2.0,
    • 3.0
  • 'fit_prior' (Whether to learn class prior probabilities or not) :
    • True,
    • False
• After the training process is complete, the best parameters obtained are as follows:
  • 'alpha': 0
  • 'fit_prior': True
• Then, the training process is conducted once again using the best parameters obtained from the hyperparameter tuning process to acquire the final model that will be used in the subsequent stages.
• The final model is saved to models/MNB_Model.sav

Model Evaluation

• Confusion Matrix:

  

• Accuracy Score:

  • 0.8655172413793103 (86.55%)

• Precision:

  • 0.8655172413793103 (86.55%)

• Recall:

  • 0.8655172413793103 (86.55%)

• F1 Score:

  • 0.8655172413793104 (86.55%)

Contributors

• Linggar Maretva Cendani - [email protected]

License

This project is licensed under the MIT License - see the LICENSE file for details