🍽️ Restaurant Review Sentiment Analysis

🧠 Project Overview

This project performs sentiment analysis on restaurant reviews to classify customer feedback as positive or negative.
It uses TF-IDF vectorization for text-to-numeric conversion and multiple machine learning classifiers with hyperparameter tuning (GridSearchCV) for model optimization.
A Streamlit frontend is also developed for real-time user interaction and sentiment prediction.

---

📂 Dataset

• Name: Restaurant_Reviews.tsv
• Source: Restaurant customer feedback dataset
• Format: Tab-separated file (.tsv)
• Columns:
  • Review → Customer review text
  • Liked → Target label (1 = Liked, 0 = Not Liked)

---

⚙️ Key Features

✅ Text preprocessing with NLTK (tokenization, lemmatization, stopword removal)
✅ Feature extraction using TF-IDF Vectorizer
✅ Training with multiple ML classifiers
✅ Hyperparameter tuning using GridSearchCV
✅ AUC, ROC, Bias–Variance analysis
✅ Visualization of model performance
✅ Streamlit frontend for live sentiment prediction
✅ Models and vectorizer saved using Pickle

---

🧩 Machine Learning Models Used

1. Logistic Regression
2. Naive Bayes
3. Support Vector Machine (SVM)
4. Random Forest
5. Decision Tree
6. K-Nearest Neighbors (KNN)
7. Gradient Boosting
8. XGBoost
9. LightGBM

Each model undergoes GridSearchCV for best hyperparameter tuning.

---

🚨 Problem Encountered: Underfitting

Initially, the models suffered from underfitting, resulting in low accuracy and poor generalization.
To fix this issue:

• Increased TF-IDF vectorizer features (max_features=2000)
• Performed hyperparameter tuning for all models
• Evaluated bias–variance tradeoff to ensure better balance

After optimization, model performance significantly improved, achieving higher test accuracy and stable ROC curves.

---

📊 Evaluation Metrics

Each model is evaluated using:

• Accuracy
• Precision
• Recall
• F1 Score
• AUC Score
• Bias
• Variance

Visualizations:

• Accuracy comparison bar chart
• ROC curve comparison

---

🧠 Best Model

After tuning and evaluation, the top-performing model is selected automatically based on highest test accuracy.
Both the best model and TF-IDF vectorizer are saved using pickle for later use in the Streamlit frontend.

---

🏆 Model Comparison Results

Rank	Model	Best Params	Train Acc	Test Acc	Precision	Recall	F1 Score	AUC	Bias	Variance
1	SVM	{'C': 10, 'kernel': 'rbf'}	0.9988	0.9750	0.9594	0.9895	0.9742	0.9978	0.0012	0.0238
2	Random Forest	{'max_depth': None, 'n_estimators': 200}	0.9988	0.9750	0.9689	0.9791	0.9740	0.9861	0.0012	0.0238
3	Logistic Regression	{'C': 10, 'solver': 'lbfgs'}	0.9944	0.9675	0.9450	0.9895	0.9668	0.9943	0.0056	0.0269
4	Decision Tree	{'criterion': 'entropy', 'max_depth': None}	0.9988	0.9600	0.9397	0.9791	0.9590	0.9608	0.0012	0.0388
5	KNN	{'n_neighbors': 5, 'weights': 'distance'}	0.9988	0.9400	0.9034	0.9791	0.9397	0.9755	0.0012	0.0588
6	Naive Bayes	{'alpha': 0.1}	0.9750	0.9300	0.9137	0.9424	0.9278	0.9842	0.0250	0.0450
7	Gradient Boosting	{'learning_rate': 0.2, 'n_estimators': 200}	0.9838	0.9100	0.9429	0.8639	0.9016	0.9786	0.0162	0.0737
8	XGBoost	{'learning_rate': 0.2, 'max_depth': 7, 'n_estimators': 200}	0.9606	0.8875	0.9148	0.8429	0.8774	0.9537	0.0394	0.0731
9	LightGBM	{'learning_rate': 0.05, 'max_depth': 10, 'n_estimators': 200, 'num_leaves': 31}	0.8194	0.7575	0.8092	0.6440	0.7172	0.8313	0.1806	0.0619

---