🎓 CIFAR-10 Image Classification System

A complete Machine Learning course project that classifies images from the CIFAR-10 dataset into 10 categories using three easy-to-explain algorithms:

• Logistic Regression
• Random Forest
• Support Vector Machine (SVM)

The project includes:

• ✅ Full ML pipeline (preprocessing, PCA, training, evaluation)
• ✅ Model comparison with charts and analysis
• ✅ Flask web app with upload & classify functionality
• ✅ Streamlit deployment file
• ✅ Source code ready for VS Code / Google Colab

---

📂 Project Structure

image-classification-project/
├── app.py                  # Flask web application
├── train_models.py         # ML training pipeline
├── streamlit_app.py        # Streamlit deployment file
├── requirements.txt        # Python dependencies
├── models/                 # Saved trained models
├── static/
│   ├── charts/             # Accuracy & confusion matrix charts
│   ├── css/style.css       # Website styling
│   ├── js/main.js          # Frontend JavaScript
│   └── uploads/            # Uploaded images
├── templates/              # HTML pages
└── README.md               # This file

---

🛠️ Step-by-Step Setup (VS Code)

Step 1: Install Python

Download and install Python 3.10 or higher from python.org.

Step 2: Open Project in VS Code

1. Open VS Code.
2. Go to File → Open Folder and select image-classification-project.

Step 3: Install Dependencies

Open terminal in VS Code (Ctrl + `) and run:

pip install -r requirements.txt

Step 4: Train the Models

Run the training script:

python train_models.py

This will:

• Download CIFAR-10 dataset
• Preprocess the images
• Train 3 models
• Save models in /models
• Generate charts in /static/charts

Step 5: Run the Flask Website

python app.py

Open your browser and go to:

http://127.0.0.1:5000

You can now upload an image and get predictions!

---

☁️ Run in Google Colab

1. Go to colab.research.google.com
2. Upload all project files
3. Install dependencies:

!pip install -r requirements.txt

4. Train models:

!python train_models.py

5. Run Flask (use a tunnel like ngrok or run Streamlit instead):

!python app.py

Recommended for Colab: Use streamlit_app.py (see below).

---

🚀 Deploy to Streamlit Cloud

Step 1: Push Code to GitHub

1. Create a GitHub repository.
2. Upload all project files.
3. Make sure streamlit_app.py is in the root of the repository.

Step 2: Go to Streamlit Cloud

Visit https://streamlit.io/cloud and sign in with GitHub.

Step 3: Create New App

1. Click "New app".
2. Select your GitHub repository.
3. Set Main file path to: streamlit_app.py
4. Click Deploy.

Step 4: Your URL

After deployment, your website will be live at:

https://yourname-cifar10-classifier.streamlit.app/

You can change yourname-cifar10-classifier to whatever you want.

⚠️ Important: Streamlit Cloud free tier has limited memory. The first time you open the app, it may need to download CIFAR-10 and load models, which can take 1-2 minutes.

---

🧠 Understanding the Models (Easy to Explain)

1. Logistic Regression

Think of it as drawing a straight line (or plane) that separates different image classes. It predicts the probability that an image belongs to each class.

2. Random Forest

It creates many small "decision trees" and combines their answers. Each tree asks simple yes/no questions about pixel values, and the forest votes on the final class.

3. Support Vector Machine (SVM)

SVM finds the best boundary (margin) that separates classes. It tries to keep the gap between classes as wide as possible, which helps it generalize better.

---

📊 Results

Model	Accuracy	Precision	Recall	F1-Score
Logistic Regression	39.33%	38.86%	39.33%	38.95%
Random Forest	38.63%	38.03%	38.63%	37.88%
Support Vector Machine	48.90%	48.72%	48.90%	48.62%

Conclusion: SVM performed best among the three traditional ML models.

---

📝 For Your Report & Presentation

You can include:

• Problem: Classify 32×32 RGB images into 10 classes.
• Dataset: CIFAR-10 (60,000 images).
• Preprocessing: Normalization, standardization, PCA.
• Models: Logistic Regression, Random Forest, SVM.
• Evaluation: Accuracy, Precision, Recall, F1-Score, Confusion Matrix.
• Best Model: SVM with ~49% accuracy.

---

📬 Need Help?

If you get any errors:

1. Make sure all files are in the same folder.
2. Run pip install -r requirements.txt again.
3. Check that train_models.py created the /models folder.
4. For Streamlit Cloud, check the app logs for missing files.

---

Made with ❤️ for Machine Learning Course Project 2025.