CMS Dimuon Signal vs Background — ML Classifier + Streamlit Demo

This project builds an end-to-end machine learning prototype using real CMS Open Data from the CERN Open Data Portal.
It classifies dimuon collision events as Z→μμ signal-like or background and provides an interactive Streamlit web interface for exploration.

---

Project Goal

• Use real CMS dimuon collision data
• Engineer physics-inspired features
• Train a classifier to separate signal vs background
• Evaluate using confusion matrix, ROC, and precision–recall
• Deploy as an interactive Streamlit app

---

Data

• Source: CERN Open Data Portal (CMS dimuon dataset)
• File used: Dimuon_DoubleMu.csv (loaded locally via the Streamlit uploader)
• Each row = one collision event with two reconstructed muons
• Labeling (physics-motivated):
  signal = 1 if 80 ≤ M ≤ 100 GeV, else background = 0
  (Z-boson peak region)

---

Methods

• Feature engineering:
  • momentum magnitude: p = sqrt(px^2 + py^2 + pz^2)
  • transverse energy: ET = sqrt(px^2 + py^2)
  • total transverse momentum: pt_sum = pt1 + pt2
• Model: RandomForestClassifier
• Evaluation:
  • confusion matrix + classification report
  • ROC curve (AUC ~ 0.999)
  • precision–recall curve
• Explainability:
  • feature_importances_ (top features dominated by pt_sum, pt1, pt2, ET1, ET2)

---

Sample Output Images

Streamlit Demo

ROC Curve

Precision–Recall Curve

Invariant Mass (Z peak)

---

How to Run

Copy code:

streamlit run app.py

and Upload the dataset Upload Dimuon_DoubleMu.csv from CERN Open Data.

Notes This is a demonstration prototype showing how physics intuition and machine learning can be integrated for particle physics event classification. It is designed for learning and portfolio demonstration.