Custom Neural Network for Digit Classification

🎉 I built a Neural Network from scratch using only NumPy — no TensorFlow, no PyTorch! Over the past few days, I challenged myself to truly understand how neural networks work under the hood by implementing one from scratch, without relying on any ML libraries. 🧠 What it does: ➤ Classifies handwritten digits (0–9) from 28x28 grayscale images ➤ Dataset: 42,000+ labeled images ➤ Achieved 87.4% accuracy using a fully custom-built neural net 🔧 Built using: ➤ Python, NumPy, Pandas ➤ ReLU activations ➤ Softmax output + cross-entropy loss ➤ Mini-batch training ➤ Forward & backpropagation coded manually 💡 Key learnings: ✅ Matrix calculus & weight updates ✅ Training dynamics (learning rate, loss curves, etc.) ✅ Debugging accuracy plateaus & improving performance 🚀 Next steps: ➤ Try Adam optimizer ➤ Expand to EMNIST or Fashion-MNIST ➤ Build a web app where users can draw digits Would love to connect with others working on ML from the ground up! #MachineLearning #NeuralNetworks #Python #NumPy #DeepLearning #StudentProject #FromScratch #AI #LinkedInProjects #OpenSource