Breast Tumor Diagnosis System

A machine-learning web application that classifies a breast tumour as benign or malignant from 30 diagnostic features, with a web interface for clinicians' inputs and a REST API for integration.

Diagnostic features extracted from tumour images — radius, texture, perimeter, area, smoothness, compactness, concavity, concave points, symmetry and fractal dimension, each as a mean, a standard error and a worst value — need to be turned into a transparent benign/malignant prediction with a probability for each class, not a bare label.

A Flask API loads the serialised model and scaler at start-up and exposes GET /health (model and scaler status) and POST /predict, which validates all 30 features and returns the class, a confidence value and both class probabilities. The web interface is plain HTML, CSS and JavaScript with benign, malignant and random sample inputs for testing.

The model is a K-nearest-neighbours classifier selected by hyper-parameter search (k = 3, uniform weights, Manhattan distance), trained on the Breast Cancer Wisconsin (Diagnostic) data set — 569 samples, 30 features, 357 benign and 212 malignant. Inputs are standardised with the same scaler used in training before inference.

As reported in the repository for the held-out test set: 96.49 % accuracy, 91.67 % precision, 97.06 % recall, 94.29 % F1 and 98.97 % ROC-AUC (98.02 % training accuracy). These are offline evaluation figures on a public research data set — not clinical validation.