How almansur123/breast-cancer-detection works
Project type and purpose
• This is a machine-learning research project (not a web app or microservice).
• Its goal is to classify tumors in the Wisconsin Breast Cancer dataset, using preprocessing, feature selection, training multiple models, then explaining predictions with SHAP and LIME.
Main components
a. Data Ingestion
• Wisconsin Breast Cancer dataset (CSV)
• Source: local file or remote URL
b. Data Preprocessing
• Missing-value handling, encoding, scaling
• Output: cleaned DataFrame
c. Feature Selection
• Techniques: filter methods, wrapper methods
• Output: reduced feature set
d. Model Training
• Sub-components:
– Random Forest classifier (final best model, 97.37% accuracy)
– Other candidate models (Logistic Regression, SVM, etc.)
e. Model Evaluation
• Metrics: accuracy, precision, recall, ROC-AUC
• Outputs: performance reports, confusion matrix
f. Model Explanation
• SHAP explainer module
• LIME explainer module
• Outputs: feature-importance plots, local explanations
g. (Future) Deployment & Extension
• Deep-learning module (e.g., TensorFlow/Keras pipeline)
• Additional data sources (patient history, imaging)
• Model registry or serving endpoint
Relationships & data flow
• A directional flow: Data Source → Preprocessing → Feature Selection → Model Training → Evaluation → Explanation
Architectural patterns / design principles
• Pipeline pattern: stages connected in series
• Separation of concerns: data prep vs. modeling vs. interpretation
• Modularity: each stage is a separate component or script/notebook cell
Elements
• Compute tasks or scripts (rounded rectangles)
• Models (parallelograms or hexagons)