📝 Machine Learning Assignment No. 2

📋 Assignment Details

Property	Value
Name	Ahsan Umar
Roll No	232552
Course	Machine Learning
Assignment No	2
GitHub Repository	codewithdark-git/USA-MLAssignment
Project Directory	urdu-sentiment-analysis

---

🎬 Project Title

Urdu Movie Reviews — Sentiment Analysis

Project Description

This project implements a complete NLP + Machine Learning pipeline for classifying Urdu movie reviews as positive or negative. It covers everything from raw data loading to ensemble classifiers, with clean visualizations and a thorough comparative analysis.

Dataset: IMDB 50K Movie Reviews — Translated Urdu (Kaggle)

---

🗂️ Repository Structure

urdu-sentiment-analysis/
│
├── README.md                        ← Assignment document
├── requirements.txt                 ← All dependencies
├── run_all.py                       ← Execute all questions
│
├── data/
│   ├── imdb_urdu_reviews.csv        ← Dataset (50K reviews)
│   └── README_data.md               ← Instructions for dataset
│
├── utils/                           ← Shared utilities
│   ├── __init__.py
│   ├── urdu_stopwords.py            ← Urdu stopword list
│   └── text_utils.py                ← Reusable preprocessing functions
│
├── Q1_Data_Loading/
│   ├── README.md
│   └── q1_data_loading.py           ← Data Loading & EDA
│
├── Q2_Visualization/
│   ├── README.md
│   └── q2_wordcloud.py              ← WordCloud Visualization
│
├── Q3_Preprocessing/
│   ├── README.md
│   └── q3_preprocessing.py          ← Data Preprocessing
│
├── Q4_Feature_Extraction/
│   ├── README.md
│   └── q4_feature_extraction.py     ← N-grams & TF-IDF
│
├── Q5_Classification/
│   ├── README.md
│   └── q5_classification.py         ← ML Classification Models
│
├── Q6_Comparative_Analysis/
│   ├── README.md
│   ├── q6_comparative_analysis.py   ← Model Comparison
│   └── q6_comparison.py
│
├── Q7_Ensemble/
│   ├── README.md
│   └── q7_ensemble.py               ← Ensemble Methods
│
└── outputs/
    ├── q5_cv_results.csv            ← Cross-validation results
    └── q7_ensemble_results.csv      ← Ensemble results

---

⚡ Installation & Usage

Prerequisites

• Python 3.8+
• Git
• pip package manager

Step 1: Clone the Repository

git clone https://github.com/codewithdark-git/USA-MLAssignment.git
cd USA-MLAssignment/urdu-sentiment-analysis

Step 2: Install Dependencies

pip install -r requirements.txt

Step 3: Prepare the Dataset

1. Download from Kaggle — IMDB 50K Urdu Reviews
2. Extract and place the CSV file inside the data/ folder
3. Verify: data/imdb_urdu_reviews.csv

Step 4: Execute Questions

Option A: Run all questions sequentially

python run_all.py

Option B: Run individual questions

python Q1_Data_Loading/q1_data_loading.py
python Q2_Visualization/q2_wordcloud.py
python Q3_Preprocessing/q3_preprocessing.py
python Q4_Feature_Extraction/q4_feature_extraction.py
python Q5_Classification/q5_classification.py
python Q6_Comparative_Analysis/q6_comparative_analysis.py
python Q7_Ensemble/q7_ensemble.py

---

📦 Questions Overview

#	Topic	Key Techniques	Files
Q1	Data Loading & Preliminary Analysis	Pandas, EDA, class distribution	Q1_Data_Loading
Q2	Data Visualization	WordCloud, Matplotlib, Text analysis	Q2_Visualization
Q3	Data Preprocessing	Normalization, Tokenization, Stopwords, Stemming	Q3_Preprocessing
Q4	Feature Extraction	Unigrams, Bigrams, Trigrams, TF-IDF	Q4_Feature_Extraction
Q5	ML Classification	Naïve Bayes, SVM, Decision Tree, RF, k-NN	Q5_Classification
Q6	Comparative Analysis	Performance metrics comparison, visualizations	Q6_Comparative_Analysis
Q7	Ensemble Methods	Voting Classifier (NB + SVM + RF)	Q7_Ensemble

---

📊 ML Algorithms Used

• Naïve Bayes — Fast probabilistic baseline
• Support Vector Machine (SVM) — Strong for high-dimensional text
• Decision Tree — Interpretable, rule-based
• Random Forest — Ensemble of trees, robust
• k-Nearest Neighbors (k-NN) — Distance-based classifier
• Voting Ensemble — Combines top 3 models

---

🔑 Key Concepts

• TF-IDF transforms raw word counts into weighted relevance scores
• Stratified K-Fold ensures class balance across all folds
• N-grams capture both individual words and multi-word phrases
• Ensemble methods reduce variance and improve generalization

---

📈 Evaluation Metrics

Each model is evaluated using Accuracy, Precision, Recall, and F1 Score.

---

📚 Learning Outcomes

This assignment demonstrates proficiency in:

✅ Data Preprocessing — Handling multilingual text (Urdu)
✅ Feature Engineering — TF-IDF, N-grams, and vectorization
✅ Classification Models — Implementing and comparing multiple algorithms
✅ Model Evaluation — Cross-validation, metrics (Accuracy, Precision, Recall, F1)
✅ Ensemble Methods — Combining weak learners for improved performance
✅ Visualization — WordCloud, comparison charts, confusion matrices

---

🔗 Important Links

Link	Description
GitHub Repository	Main assignment repository
Project Source	Urdu Sentiment Analysis project
Dataset (Kaggle)	IMDB 50K Urdu Reviews
Utility Functions	Shared preprocessing utilities

---

📝 Submission Information

• Submission Type: GitHub Repository
• Repository Owner: codewithdark-git
• Assignment Type: Machine Learning (NLP + Classification)
• Status: Complete with all 7 questions implemented

---

👤 Author

Ahsan Umar
Roll No: 232552
Course: Machine Learning
Assignment: 2/2

---

📄 License & Attribution

Dataset Credit: Kaggle — akkefa
Created as part of Machine Learning coursework