Disease Outbreak Prediction

An interactive tool that lets a user enter clinical parameters and returns a risk prediction for three diseases using three separately trained scikit-learn models. Built during the AICTE / TechSaksham AI: Transformative Learning internship.

Early screening tools for chronic disease often exist as isolated notebooks that a non-technical user can never touch. The goal was to expose three well-known clinical datasets (Pima Indians, Cleveland heart, Parkinson's voice) behind one unified interface.

• 01Three separate models, each trained and evaluated in its own notebook: logistic regression for diabetes, SVM with RBF kernel for heart disease, and a tuned SVM for Parkinson's voice features.
• 02Model artifacts serialized with pickle and lazy-loaded by the Streamlit app on first request.
• 03Single Streamlit multi-page app with one page per disease, sharing common validation and result-rendering components.

• Python
• Scikit-learn
• Streamlit
• Logistic Regression
• SVM

• Three prediction workflows in one interface
• Input validation with sensible medical bounds per field
• Probability-style confidence output rather than raw class labels

• Feature scales varied wildly across datasets — solved with a StandardScaler saved alongside each model.
• Small dataset sizes made cross-validation results noisy; used stratified k-fold with fixed seeds for comparable numbers.
• Streamlit reruns on every widget change; had to be careful not to reload models on each interaction.

• Three working classifiers with acceptable accuracy across held-out splits
• Single deployable app that runs anywhere Python does
• Clean template for adding more diseases without touching the UI

• Persisting scalers next to models is not optional — training-time preprocessing must ship with inference.
• Streamlit's caching primitives are underrated for keeping demos snappy.
• Small models trained carefully beat larger models trained lazily.