💬 Sentiment Analysis with Text Classification

This project involves performing sentiment analysis on textual data using classical machine learning techniques. The primary goal is to classify text into one of four categories: positive, negative, neutral, or others, using NLP preprocessing, TF-IDF vectorization, SMOTE resampling, and classification algorithms.

---

🧠 Objective

To build a robust and efficient machine learning pipeline that can accurately detect sentiment from raw text data and evaluate various models for performance.

---

🛠️ Tools & Libraries

• Python
• pandas, numpy, matplotlib, seaborn
• sklearn, imblearn, nltk
• LazyPredict, wordcloud

---

📂 Project Structure

├── Sentiment_analysis.ipynb
├── README.md
├── Test-Set.csv
└── requirements.txt (optional)

---

📊 Dataset Overview

The dataset consists of customer feedback labeled with sentiments:

• Text – raw feedback.
• Sentiment – class label: positive, negative, neutral, others.

---

🔄 Preprocessing Steps

1. Lowercasing & Tokenization
2. Stopword Removal (using NLTK)
3. TF-IDF Vectorization
4. SMOTE Oversampling to handle class imbalance
5. Train-Test Split

---

⚙️ Modeling Workflow

🧪 Model Benchmarking (LazyPredict)

Ran LazyPredict on the processed data to benchmark 20+ models. Top performers included:

Model	Accuracy	F1 Score	Time Taken
GaussianNB	1.00	1.00	0.54 sec
SGDClassifier	1.00	1.00	1.27 sec
QuadraticDiscriminantAnalysis	1.00	1.00	3.01 sec
CalibratedClassifierCV	1.00	1.00	215.22 sec

---

✅ Final Model: Gaussian Naive Bayes

🔍 Hyperparameter Tuning

Performed GridSearchCV to optimize var_smoothing parameter.

grid_params = {'var_smoothing': [1e-9, 1e-8, 1e-7, 1e-6]}

📈 Final Evaluation Metrics

• Accuracy: 99.6%
• Macro F1-Score: 1.00

Sentiment	Precision	Recall	F1-Score	Support
Negative	1.00	1.00	1.00	139
Neutral	0.99	1.00	1.00	132
Others	1.00	0.99	0.99	137
Positive	0.99	1.00	1.00	124

---

☁️ Word Cloud Visualization

A word cloud was generated to identify the most frequent terms used across the text data.

---

🗣️ Top Words Per Sentiment

Using collections.Counter, the top 10 most frequent words were extracted for each sentiment class.

---

📌 Key Insights

• Gaussian Naive Bayes outperformed other models with high accuracy and F1 scores.
• TF-IDF + SMOTE preprocessing greatly improved classification consistency.
• Most frequent sentiment-indicative words align well with human interpretation.
• Efficient, interpretable, and high-performing baseline model for sentiment detection.

---

🔗 Connect with Me

Let’s connect if you’re working on:

• Customer feedback or retention
• Applied ML/NLP projects
• Or just want to chat about data science!

📬 ([https://www.linkedin.com/in/richard-olanite-55b4b0241/])