Ensemble-NB-SVM-Model-Comparison-Text-Classification

Goal: Comparing effectiveness of boosting/bagging ensemble model implementations using two metrics (speed and accuracy score) and comparing discovered best model against multinomial naive bayes and support vector classifier models - algorithms more conventionally fit for text classification.

Libraries:

• Scikit-Learn
• XGBoost
• CatBoost

Models:

• Scikit-learn:
  • Decision Tree Classifier (Baseline)
  • Random Forest Classifier
  • Naive Bayes Classifier
  • AdaBoost Classifier
  • Gradient Boosting Classifier
  • Histogram Gradient Boosting
• XGBoost Classifier
• CatBoost Classifier

Data Source: - https://ai.stanford.edu/~amaas/data/sentiment/

• Review data read in as .txt segmented by positive (< 5) and negative (> 5) ratings and mapped to ratings nested in filenames
• Text reviews TFIDF vectorized yielding sparse matrix bag of words
• Data cleaned and split into train, test, and validation sets
• Each model implementation built; hyperparameters set to ensure performance fairness/consistency
• Models fit on training sets (with optional validation)
• Accuracy and runtime reported for each model on test set performance