Key Contributions & Features

• Developed an ML model to identify industrial assets at high risk of failure using sensor data.
• Trained a PyTorch neural network, achieving 97% ROC-AUC on the test set.
• Implemented dropout and batch normalization layers to prevent overfitting.
• Addressed target leakage by identifying and dropping confounding features from the dataset.
• Used a Scikit-learn pipeline for one-hot encoding and robust feature scaling.
• Eliminated training-serving skew by serializing the entire preprocessing pipeline with joblib.
• Developed a Flask API to serve real-time predictions from the trained model.
• Containerised the application with Docker for reproducible deployment.
• Deployed the container to Google Cloud Run, making the project publicly accessible.
• Optimised API response time by pre-loading model artifacts at server startup.
• Created a simple web UI with HTML and JavaScript for live, interactive model testing.