Higgs Boson Signal Classification

This project focuses on identifying the Higgs boson, a fundamental particle that gives mass to other elementary particles, through its decay into two Z bosons, which subsequently decay into four leptons (H → ZZ → 4ℓ). This decay channel is often called the 'golden channel' due to its clean experimental signature, but it is extremely rare and buried under significant background processes.

Using a dataset of 218,632 simulated events based on CMS Open Data, our team constructed a binary classification pipeline to distinguish the Higgs signal from three primary background processes: ZZ → 4e, ZZ → 2e2μ, and ZZ → 4μ. The key input features for the model were the reconstructed mass of the Higgs candidate, and the masses of the two Z bosons.

The core of the project was a custom Feed-Forward Deep Neural Network (DNN) implemented in PyTorch. The architecture featured a Linear input layer, a ReLU activation hidden layer, and a Sigmoid output layer for binary classification. The model was trained using Binary Cross-Entropy Loss and the Adam optimizer over 1000 epochs, achieving a highly successful 97.3% training accuracy and 97.1% validation accuracy, demonstrating strong generalization without overfitting.

Performance evaluation involved comprehensive analysis including Receiver Operating Characteristic (ROC) curves with an Area Under Curve (AUC) of ~0.99, demonstrating exceptional signal-to-background discrimination. The project culminated in a detailed presentation and poster summarizing the methodology and physics implications.