Update README.md

README.md

@@ -1,67 +1,128 @@
-# Handwritten Digit Recognition from Scratch
+# Digit Recognition using Neural Network (From Scratch)
 
-A neural network built from **scratch using pure Python** to recognize MNIST handwritten digits with **95-97% accuracy** — no NumPy, TensorFlow, or PyTorch!
+A Python implementation of a handwritten digit recognition system built entirely from scratch. This project focuses on understanding the mathematical foundations of neural networks without relying on high-level deep learning frameworks.
 
-## 🎯 Features
+The goal is educational clarity, architectural transparency, and strong fundamentals in machine learning implementation.
 
-- ✅ Pure Python implementation (no ML libraries)
-- ✅ Custom matrix operations from scratch
-- ✅ Forward & backward propagation
-- ✅ Mini-batch gradient descent
-- ✅ 95-97% test accuracy on MNIST
+---
+
+## Project Overview
+
+This repository implements a fully connected neural network to classify handwritten digits from the MNIST dataset. All core components such as forward propagation, backpropagation, weight updates, and data handling are manually implemented.
+
+This project demonstrates:
+
+* Manual neural network implementation
+* Matrix-based forward and backward propagation
+* Gradient descent optimization
+* Dataset handling and preprocessing
+* Modular and readable Python structure
+
+---
+
+## Project Structure
 
-## 🏗️ Architecture
 ```
-Input (784) → Hidden (128, ReLU) → Output (10, Softmax)
+digitrecognition/
+│
+├── data_loader.py              # Loads and preprocesses dataset
+├── download_mnist.py           # Downloads MNIST dataset
+├── find_data.py                # Utility to locate dataset files
+├── my_math.py                  # Custom mathematical helper functions
+├── neural_network_scratch.py   # Core neural network implementation
+├── show_structure.py           # Displays model architecture details
+├── train.py                    # Training script
+├── test_single.py              # Test prediction for a single sample
+├── README.md
+└── .gitignore
 ```
 
-## 🚀 Quick Start
+---
+
+## Technologies Used
+
+* **Language:** Python
+* **Numerical Computation:** NumPy
+* **Dataset:** MNIST
+
+No high-level ML frameworks such as TensorFlow or PyTorch are used. All learning logic is implemented manually.
+
+---
+
+## How It Works
+
+1. MNIST dataset is downloaded and prepared.
+2. Data is normalized and structured for training.
+3. A feedforward neural network performs prediction.
+4. Backpropagation computes gradients.
+5. Weights are updated using gradient descent.
+
+---
+
+## Setup Instructions
+
+### 1. Clone the Repository
+
 ```bash
-# Clone repository
 git clone https://github.com/kaushalrog/digitrecognition.git
 cd digitrecognition
+```
+
+### 2. Install Dependencies
+
+```bash
+pip install numpy
+```
 
-# Install matplotlib (only dependency)
-pip install matplotlib
+### 3. Download Dataset
 
-# Download MNIST data
+```bash
 python download_mnist.py
+```
+
+### 4. Train the Model
 
-# Train model
+```bash
 python train.py
 ```
 
-## 📊 Results
+### 5. Test a Single Sample
 
-- **Training Accuracy**: ~96-98%
-- **Test Accuracy**: ~95-97%
-- **Training Time**: ~5-10 minutes (20 epochs)
+```bash
+python test_single.py
+```
 
-## 📁 Files
+---
 
-- `my_math.py` - Custom matrix operations
-- `data_loader.py` - MNIST loader
-- `neural_network_scratch.py` - Neural network
-- `train.py` - Training script
+## Learning Objectives
 
-## 🧠 What's Inside
+This project is ideal for understanding:
 
-Built from scratch:
-- Matrix class with all operations
-- ReLU & Softmax activations
-- Cross-entropy loss
-- Backpropagation algorithm
-- He weight initialization
-- Mini-batch gradient descent
+* Neural network internals
+* Backpropagation mathematics
+* Gradient descent optimization
+* Matrix operations in ML
+* Building ML systems without frameworks
 
-## 👨‍💻 Author
+---
 
-**Kaushal** - [GitHub](https://github.com/kaushalrog)
+## Future Improvements
 
-## 📝 License
+* Add configurable hyperparameters
+* Implement mini-batch gradient descent
+* Add performance metrics visualization
+* Export trained model weights
+* Build a simple web interface for predictions
 
-MIT License
+---
+
+## License
+
+This project is licensed under the MIT License.
 
 ---
 
-⭐ Star if you found this helpful!
+## Author
+
+Kaushal S
+GitHub: h