MNIST Digit Recognition
Class-wise F1-score Distribution
Mnist Digit Recognition Home Page
Mnist Digit Recognition Hand Drawn Digit
Mnist Digit Recognition Prediction
Performance Across Different Models
Heatmap of classwise F1-score
Delta from best classwise F1-score
Class-wise F1-score Distribution
Mnist Digit Recognition Home Page
Project Description
An interactive web application where users can draw handwritten digits and get real-time predictions from a trained ML model using the MNIST dataset.
Responsibilities
- Developed an end-to-end handwritten digit recognition system using the MNIST dataset.
- Achieved 98.16% accuracy and a 0.98 weighted F1-score using a Support Vector Classifier (SVC).
- Built a FastAPI backend to serve real-time predictions from the trained ML model.
- Designed an interactive Next.js frontend where users can draw digits and see predictions instantly.
- Hosted the application using Firebase App Hosting with model inference served via Firebase Cloud Functions.
- Explored multiple ML models (Logistic Regression, SVC, Random Forest, KNN) and optimized for best performance.
- Debugged and resolved issues related to preprocessing pipelines and production model deployment, ensuring a smooth ML workflow.
Related Links
Technology
FastAPIFirebaseNext.jsPythonScikit-Learn