Skip to content

README_MODELS.md

Latest commit

 

History

History
150 lines (112 loc) · 5.46 KB

README_MODELS.md

File metadata and controls

150 lines (112 loc) · 5.46 KB

Model Scripts for Amazon Review Classifier

This directory contains individual R scripts for each machine learning model used to classify fake Amazon reviews.

Prerequisites

Make sure you have installed all required R packages:

install.packages(c("tm", "magrittr", "SnowballC", "Matrix", "glmnet", "pROC", "caret", 
                   "randomForest", "ranger", "xgboost", "keras3", "ade4", 
                   "dplyr", "ggplot2", "gridExtra"))

For keras3 (required for neural network models 7-8):

Step 1: Install keras3 package

install.packages("keras3")

If CRAN installation fails, try installing from GitHub:

install.packages("remotes")
remotes::install_github("rstudio/keras3")

Step 2: TensorFlow backend installation

After installing keras3, TensorFlow will be automatically installed when you first run:

library(keras3)
use_backend("tensorflow")

This happens automatically when you run the neural network model scripts.

Note:

  • The install_keras() function is not available in keras3
  • TensorFlow installation is handled automatically when you first use use_backend("tensorflow")
  • If you encounter issues, see INSTALL_NEURAL_NETWORK.md for detailed troubleshooting steps

Data File

All scripts expect data_new.csv to be in the same directory. This file contains the preprocessed Amazon reviews data.

Model Scripts

1. model_01_glm_lasso_min.R

  • Model: Logistic Regression with Lasso Regularization (lambda.min)
  • Features: Selected features from Lasso regularization
  • Output: Accuracy, Confusion Matrix, Sensitivity, Specificity, Precision, F1 Score, AUC

2. model_02_glm_lasso_1se.R

  • Model: Logistic Regression with Lasso Regularization (lambda.1se)
  • Features: More conservative feature selection
  • Output: Accuracy, Confusion Matrix, Sensitivity, Specificity, Precision, F1 Score, AUC

3. model_03_random_forest.R

  • Model: Random Forest
  • Parameters: ntree=100, nodesize=5, mtry=60
  • Output: Accuracy, Confusion Matrix, Sensitivity, Specificity, Precision, F1 Score, AUC, Variable Importance

4. model_04_ranger.R

  • Model: Ranger (Fast Random Forest)
  • Parameters: num.trees=200
  • Output: Accuracy, Confusion Matrix, Sensitivity, Specificity, Precision, F1 Score, AUC, Variable Importance

5. model_05_xgboost.R

  • Model: XGBoost (Gradient Boosting)
  • Parameters: eta=0.05, nrounds=342
  • Output: Accuracy, Confusion Matrix, Sensitivity, Specificity, Precision, F1 Score, AUC, Feature Importance

6. model_06_simple_glm.R

  • Model: Simple Logistic Regression
  • Features: Only VERIFIED_PURCHASE
  • Output: Accuracy, Confusion Matrix, Sensitivity, Specificity, Precision, F1 Score, AUC, Model Coefficients

7. model_07_neural_network.R

  • Model: Neural Network (Keras3)
  • Features: Review text features only
  • Architecture: 2 hidden layers (32-32 units), dropout=0.3
  • Output: Test Loss, Test Accuracy, Confusion Matrix, Sensitivity, Specificity, Precision, F1 Score, AUC
  • Saves: Model to fakereview.keras

8. model_08_neural_network_titles.R

  • Model: Neural Network (Keras3)
  • Features: Review text + Review title features
  • Architecture: 3 hidden layers (32-32-16 units), dropout=0.3
  • Output: Test Loss, Test Accuracy, Confusion Matrix, Sensitivity, Specificity, Precision, F1 Score, AUC
  • Saves: Model to fakereview_titles.keras

Usage

To run any model, simply execute the R script:

source('model_01_glm_lasso_min.R')

Or from command line:

Rscript model_01_glm_lasso_min.R

Shared Data Preprocessing

All models use data_preprocessing.R which:

  • Loads the data from data_new.csv
  • Preprocesses text data (removes stopwords, punctuation, numbers, stems)
  • Creates document-term matrix
  • Splits data into training (75%) and test (25%) sets

Performance Metrics

Each model outputs:

  • Accuracy: Overall classification accuracy
  • Confusion Matrix: True/False Positives and Negatives
  • Sensitivity (Recall): True Positive Rate
  • Specificity: True Negative Rate
  • Precision: Positive Predictive Value
  • F1 Score: Harmonic mean of Precision and Recall
  • AUC: Area Under the ROC Curve

Visualizations

All models automatically generate and save visualizations to the figures/ directory:

  • ROC Curves: Receiver Operating Characteristic curves for all models
  • Confusion Matrix Heatmaps: Visual representation of classification results
  • Model-Specific Visualizations:
    • Lasso models (models 1-2): Cross-validation plots showing lambda selection
    • Random Forest (model 3): Error rate plot and variable importance
    • Ranger (model 4): Variable importance plot
    • XGBoost (model 5): Feature importance plot
    • Simple GLM (model 6): Prediction probability distribution
    • Neural Networks (models 7-8): Training history plots (loss and accuracy) and prediction probability distributions

All figures are saved as high-resolution PNG files (300 DPI) with descriptive filenames following the pattern model_XX_description.png.

Notes

  • All models use the same random seed (245) for reproducibility
  • Training/test split is 75/25
  • Neural network models may take several minutes to train
  • Some models save their trained models to disk for later use
  • All visualizations are automatically saved to the figures/ directory when you run each model script
  • The figures/ directory will be created automatically if it doesn't exist