Fake Review Detector

by Muhammad Aminu Umar(WebCodeLab)

A complete end-to-end machine learning project to detect fake product reviews using natural language processing (NLP) and classical classification models.

This tool enables:

• Training and testing of a fake/real review classifier
• Real-time prediction for single reviews
• Batch prediction from CSV uploads
• Saving/loading model and vectorizer
• Web interface using Gradio
• Model explainability with ELI5

---

Features Overview

Data Processing

• Predefined sample dataset of labeled real/fake reviews
• Supports training/test splitting and CSV input/output

NLP + Vectorization

• Texts are preprocessed and vectorized using TF-IDF
• Model-ready format with scikit-learn

Classifier Training

• Supports:
  • Logistic Regression (default)
  • Random Forest (optional)
• Easily extendable to other models

Model Evaluation

• Outputs: Accuracy, Precision, Recall, F1
• Displays confusion matrix and classification report

Single Prediction

• Predict whether a single user review is real or fake
• Can be used directly in code or via Gradio UI

Batch Prediction

• Upload CSV file of reviews
• Returns labeled predictions
• Saves results in a new batch_predictions.csv

Save/Load Model

• Model and vectorizer are saved using joblib
• Reloadable for future use without retraining

Gradio Web Interface

• Simple browser UI to test reviews manually
• Optional CSV upload directly from UI

Explainability (ELI5)

• Highlight influential words in prediction
• View which tokens drive fake/real classification