OncoScan

Breast Cancer Prediction Project – Key Highlights

• Objective:

  To assist medical professionals in predicting the malignancy of breast tumors using an explainable machine learning model.

• Model & Accuracy:

  Built and trained a Support Vector Classifier (SVC) model with 96.5% accuracy on clinical diagnostic data.

• Data & Preprocessing:

  • Utilized features extracted from digitized breast mass images.

  • Applied StandardScaler for normalization.

  • Modularized preprocessing through a dedicated data_cleaning pipeline.

• User Interface:

  • Developed an interactive web app using Streamlit.

  • Sidebar allows clinicians to input diagnostic values via sliders.

  • Real-time output: Benign or Malignant prediction, with summary table.

• Model Deployment:

  • Serialized the model and scaler using Pickle for reuse.

  • Implemented prediction logic and UI in separate, modular scripts.

• Explainability (Planned):

  Integration of SHAP values to interpret the model’s predictions and visualize feature importance.

• Tools & Technologies:

  Python · scikit-learn · Pandas · NumPy · Streamlit · SHAP · Pickle